ArticlePDF Available

An interoperable and self-adaptive approach for SLA-based service virtualization in heterogeneous Cloud environments


Abstract and Figures

Cloud computing is a newly emerged computing infrastructure that builds on the latest achievements of diverse research areas, such as Grid computing, Service-oriented computing, business process management and virtualization. An important characteristic of Cloud-based services is the provision of non-functional guarantees in the form of Service Level Agreements (SLAs), such as guarantees on execution time or price. However, due to system malfunctions, changing workload conditions, hard- and software failures, established SLAs can be violated. In order to avoid costly SLA violations, flexible and adaptive SLA attainment strategies are needed. In this paper we present a self-manageable architecture for SLA-based service virtualization that provides a way to ease interoperable service executions in a diverse, heterogeneous, distributed and virtualized world of services. We demonstrate in this paper that the combination of negotiation, brokering and deployment using SLA-aware extensions and autonomic computing principles are required for achieving reliable and efficient service operation in distributed environments.
Content may be subject to copyright.
An Interoperable and Self-adaptive Approach
for SLA-based Service Virtualization
in Heterogeneous Cloud Environments
A. Kertesza, G. Kecskemetia, I. Brandicb
aMTA SZTAKI, H-1518 Budapest, P.O. Box 63, Hungary
bVienna University of Technology, 1040 Vienna, Argentinierstr. 8/181-1, Austria
Cloud computing is a newly emerged computing infrastructure that builds on the latest achievements of diverse research
areas, such as Grid computing, Service-oriented computing, business process management and virtualization. An im-
portant characteristic of Cloud-based services is the provision of non-functional guarantees in the form of Service Level
Agreements (SLAs), such as guarantees on execution time or price. However, due to system malfunctions, changing work-
load conditions, hard- and software failures, established SLAs can be violated. In order to avoid costly SLA violations,
flexible and adaptive SLA attainment strategies are needed. In this paper we present a self-manageable architecture
for SLA-based service virtualization that provides a way to ease interoperable service executions in a diverse, heteroge-
neous, distributed and virtualized world of services. We demonstrate in the paper that the combination of negotiation,
brokering and deployment using SLA-aware extensions and autonomic computing principles are required for achieving
reliable and efficient service operation in distributed environments.
Keywords: Cloud computing, Service virtualization, SLA negotiation, Service Brokering, On-demand deployment,
1. Introduction
The newly emerging demands of users and researchers
call for expanding service models with business-oriented
utilization (agreement handling) and support for human-
provided and computation-intensive services [6]. Though
Grid Computing [21] has succeeded in establishing pro-
duction Grids serving various user communities, and both
Grids and Service Based Applications (SBAs) already pro-
vide solutions for executing complex user tasks, they are
still lacking non-functional guarantees. Providing guaran-
tees in the form of Service Level Agreements (SLAs) is also
highly studied in Grid Computing [30, 4, 9], but they have
failed to be commercialized and adapted for the business
Cloud Computing [6] is a novel infrastructure that fo-
cuses on commercial resource provision and virtualization.
These infrastructures are also represented by services that
are not only used but also installed, deployed or replicated
with the help of virtualization. These services can appear
in complex business processes, which further complicates
the fulfillment of SLAs. For example, due to changing
components, workload and external conditions, hardware-,
Email addresses: (A. Kertesz), (G. Kecskemeti), (I. Brandic)
and software failures, already established SLAs may be vi-
olated. Frequent user interactions with the system during
SLA negotiation and service executions (which are usu-
ally necessary in case of failures), might turn out to be
an obstacle for the success of Cloud Computing. Thus,
there is demand for the development of SLA-aware Cloud
middleware, and application of appropriate strategies for
autonomic SLA attainment. Despite business-orientation,
the applicability of Service-level agreements in the Cloud
field is rarely studied, yet [41]. Most of the existing works
address provision of SLA guarantees to the consumer and
not necessarily the SLA-based management of loosely cou-
pled Cloud infrastructure. In such systems, it is hard to
react to unpredictable changes and localise, where the fail-
ures have happen exactly, what is the reason for the failure
and which reaction should be taken to solve the problem.
Such systems are implemented in a proprietary way, mak-
ing it almost impossible to exchange the components (e.g.
use another version of the broker).
Autonomic Computing is one of the candidate tech-
nologies for the implementation of SLA attainment strate-
gies. Autonomic systems require high-level guidance from
humans and decide, which steps need to be done to keep
the system stable [19]. Such systems constantly adapt
themselves to changing environmental conditions. Simi-
lar to biological systems (e.g. human body) autonomic
systems maintain their state and adjust operations con-
Preprint submitted to FGCS April 6, 2012
Autonomic Manager
Analysis Planning
Monitoring Execution
Sensor Actuator
Figure 1: General architecture of an autonomic system.
sidering their changing environment. Usually, autonomic
systems comprise one or more managed elements e.g. QoS
An important characteristic of an autonomic system is
an intelligent closed loop of control. As shown in Figure
1, the Autonomic Manager (AM) manages the element’s
state and behavior. It is able to sense state changes of the
managed resources and to invoke appropriate set of actions
to maintain some desired system state. Typically control
loops are implemented as MAPE (monitoring, analysis,
planning, and execution) functions [19]. The monitor col-
lects state information and prepares it for the analysis. If
deviations to the desired state are discovered during the
analysis, the planner elaborates change plans, which are
passed to the executor. For the successful implementa-
tion of the autonomic principles to loosely coupled SLA-
based distributed system management, the failure source
should be identified based on violated SLAs, and firmly
located considering different components of the heteroge-
neous middleware (virtualization, brokering, negotiation,
etc. components). Thus, once the failure is identified, Ser-
vice Level Objectives (SLOs) can be used as a guideline
for the autonomic reactions.
In this paper we propose a novel holistic architecture
considering resource provision using a virtualization ap-
proach combined with business-oriented utilization used
for SLA agreement. Thus, we provide an SLA-coupled
infrastructure for on demand service provision based on
SLAs. First we gather the requirements of a unified service
architecture, then present our solution called SLA-based
Service Virtualization (SSV) built on agreement negoti-
ation, brokering and service deployment combined with
business-oriented utilization. We further examine this ar-
chitecture and investigate, how previously introduced prin-
ciples of Autonomic Computing appear in the basic com-
ponents of the architecture in order to cope with changing
user requirements and on demand failure handling. After
presenting our proposed solution, we evaluate the perfor-
mance gains of the architecture through a more computational-
intensive biochemical use case.
The main contributions of this paper include: (i) the
presentation of the novel loosely coupled architecture for
the SLA-based Service Virtualization and on-demand re-
source provision, (ii) the description of the architecture
including meta-negotiation, meta-brokering, brokering and
automatic service deployment with respect to the princi-
ples of autonomic computing, and (iii) the evaluation of
the SSV architecture with a biochemical case study using
In the following section we summarize related works.
Then, in Section 3, we provide the requirement analysis for
autonomous behavior in the SSV architecture through two
scenarios. Afterwards, in Section 4, we introduce the SSV
architecture, while in Section 5 the autonomous operations
of the components are detailed. In Section 6 we present the
evaluation of the SSV architecture with a biochemical case
study in a heterogeneous simulation environment. Finally,
Section 7 concludes the paper.
2. Related work
Though Cloud Computing is highly studied, and a large
body of work has been done trying to define and envi-
sion the boundaries of this new area, the applicability
of Service-level agreements in the Cloud and in a unified
distributed middleware is rarely studied. The envisioned
framework in [15] proposes a solution to extend the web
service model by introducing and using semantic web ser-
vices. The need for SLA handling, brokering and deploy-
ment also appears in this vision, but they focus on using
ontology and knowledge-based approaches. Most of re-
lated works consider virtualization approaches [17, 32, 23]
without taking care of agreements or concentrate on SLA
management neglecting the appropriate resource virtual-
izations [35, 9]. Works presented in [31, 29] discuss in-
corporation of SLA-based resource brokering into existing
Grid systems, but they do not deal with virtualization.
Venugopal et al. propose a negotiation mechanism for ad-
vance resource reservation using the alternate offers pro-
tocol [40], however, it is assumed that both partners un-
derstand the alternate offers protocol. Lee et al. discusses
application of autonomic computing to the adaptive man-
agement of Grid workflows [25] with MAPE (Monitoring,
Analysis, Planning, Execution) decision making [19], but
they also neglect deployment and virtualization. The work
by Van et. al. [38] studied the applicability of the auto-
nomic computing to Cloud-like systems, but they almost
exclusively focus on virtualization issues like VM packing
and placement.
In [7], Buyya et al. suggests a Cloud federation ori-
ented, just in time, opportunistic and scalable application
services provisioning environment called InterCloud. They
envision utility oriented federated IaaS systems that are
able to predict application service behavior for intelligent
down and up-scaling infrastructures. Then, they list the
research issues of flexible service to resource mapping, user
and resource centric QoS optimization, integration with
in-house systems of enterprises, scalable monitoring of sys-
tem components. Though they address self-management
and SLA handling, the unified utilization of other dis-
tributed systems are not studied. Recent Cloud Com-
puting projects, e.g. Reservoir [33] and OPTIMIS [11],
address specific research topics like Cloud interoperability,
but they do not consider autonomous SLA management
across diverse distributed environments. Comparing the
currently available solutions, autonomic principles are not
implemented in a adequate way because of they are lack-
ing an SLA-coupled Cloud infrastructure, where failures
and malfunctions can be identified using well defined SLA
Regarding high-level service brokering, LA Grid [34]
developers aim at supporting grid applications with re-
sources located and managed in different domains. They
define broker instances, each of them collects resource in-
formation from its neighbors and save the information in
its resource repository. The Koala grid scheduler [16] was
redesigned to inter-connect different grid domains. They
use a so-called delegated matchmaking (DMM), where Ko-
ala instances delegate resource information in a peer-2-
peer manner. Gridway introduced a Scheduling Architec-
tures Taxonomy to form a grid federation [39, 24], where
Gridway instances can communicate and interact through
grid gateways. These instances can access resources be-
longing to different Grid domains. Comparing the pre-
vious approaches, we can see that all of them use high
level brokering that delegate resource information among
different domains, broker instances or gateways. These so-
lutions are almost exclusively used in Grids, they cannot
co-operate with different brokers operating in pure service-
based or Cloud infrastructures.
Current service deployment solutions do not leverage
their benefits on higher level. For example the Workspace
Service (WS) [17] as a Globus incubator project supports
wide range of scenarios involving virtual workspaces, vir-
tual clusters and service deployment from installing a large
service stack to deploy a single WSRF service if the Vir-
tual Machine (VM) image of the service is available. It is
designed to support several virtual machines. The Xeno-
Server open platform [32] is an open distributed architec-
ture based on the XEN virtualization technique aiming
at global public computing. The platform provides ser-
vices for server lookup, registry, distributed storage and a
widely available virtualization server. Also the VMPlants
[23] project proposes an automated virtual machine config-
uration and creation service which is heavily dependent on
software dependency graphs, but this project stays within
cluster boundaries.
3. Use Case and requirements for SLA-coupled au-
tonomic Cloud middleware
Deployment and runtime management in cloud-like en-
vironments aim at providing or modifying services in a
dynamic way according to temporal, spatial or semantic
requirements. Among many other purposes, it has also
strong relation with adaptation and self-management. To
gather the requirements for an autonomic service virtual-
ization environment we refer to motivating scenarios.
3.1. Requirement analysis for a transparent, autonomic
service virtualization
The aim of this section is to investigate the require-
ments of the self-management aspects of runtime manage-
ment of service virtualization. Temporal provision of ser-
vices requires certain infrastructure features that we clas-
sified into three groups. There must be
anegotiation phase where it is specified, which
negotiation protocols can be used, which kind of ser-
vice has to be invoked, what are the non-functional
conditions and constraints (temporal availability, re-
liability, performance, cost, etc.);
abrokering phase which selects available resources
that can be allocated for providing the services. These
resources can be provided in many ways: Clouds
(virtualized resources configured for a certain speci-
fication and service level guarantees), clusters or lo-
cal Grids (distributed computing power with limited
service level guarantees) or volunteer computing re-
sources (no service level guarantees at all).
Finally, during the deployment phase services have
to be prepared for use on the selected resources in
an automatic manner.
In the following we refer to two motivating scenarios
to gather requirements from, and exemplify the need for
an autonomous service virtualization solution.
Scenario 1. There are certain procedures in various ac-
tivities that may require specific services and resources in
an ad-hoc, temporal way. Generally, there is no reason to
provide these services in a static 24/7 manner with perfor-
mance guarantees, instead, these services should be cre-
ated and decommissioned in a dynamic, on-demand way
for the following reasons:
These tasks represent fundamentally different com-
putations that cannot be re-used or composed, po-
tentially not even overlapped, e.g. air tunnel simu-
lation, crash test analysis, various optimization pro-
cedures, and so on. These services must be provided
independently from each other in a well defined and
disjoint time frame.
There is no need to dedicate certain resources to
these activities as they occur rarely or at least in-
frequently. For example, resources used by the opti-
mization services, can be re-used for other purposes
if they are not required at the foreseeable future.
Scenario 2. Similarly to the previous case, if we think of
using real life products, we also face problems that require
an autonomic service management infrastructure. For ex-
ample, in a car navigation scenario there are services that
do not need to or cannot be provided in a static, perma-
nent way. Instead, these services should be created and
decommissioned in an adaptive, dynamic or on-demand
way for the following reasons:
There is no need to dedicate certain resources to
these activities as they occur rarely or at least in-
frequently. As an example this happens when an
accident causes a traffic jam that causes all the sur-
rounding navigation systems to request services from
the GPS unit provider. This situation however does
not last longer than the last car moves away from
the problematic area with the help of the unit.
It is possible that certain services are needed upon
a certain event e.g., only in case testing phase do
not complete successfully specific simulation services
have to be invoked.
As some ad-hoc services may be mobile, they cannot
be assumed constant. E.g., navigation systems offer
services within the ad-hoc network and the network
splits to two disjunct parts, then the availability of
the offered services should be ensured in both parts.
In certain cases, dynamic deployment of a service
that is not available locally may be necessary. For
example, the GPS unit is used to broadcast live in-
formation by the customer (like a radio transmission)
towards the navigation system provider that might
not have enough bandwidth to distribute the con-
tent towards its other users. Therefore the provider
deploys a repeating service right next to the source.
This service will serve the consumers and even enable
re-encoding the transmission for the different needs
of the different users, meanwhile demanding small
bandwidth and processing power from the broad-
caster itself.
4. The autonomic, SLA-based Service Virtualiza-
tion architecture
According to the requirements introduced in the previ-
ous section, here we present a unified service architecture
that builds on three main areas: agreement negotiation,
brokering and service deployment using virtualization. We
suppose that service providers and service consumers meet
on demand and usually do not know about the negotiation
protocols, document languages or required infrastructure
of the potential partners. First we introduce the general
architecture naming the novelties and open issues, then we
detail the aforementioned three main areas with respect to
the shown architecture. Figure 2 shows our proposed, gen-
eral architecture called SLA-based Service Virtualization
4.1. The SSV architecture and the roles of its components
The main components of the architecture and their
roles are gathered in Table 1.
Figure 2: SSV architecture.
Table 1: Roles in SSV architecture
Role Description
U User A person, who wants to use a ser-
MN Meta-
A component that manages
Service-level agreements. It me-
diates between the user and the
Meta-Broker, selects appropriate
protocols for agreements; ne-
gotiates SLA creation, handles
fulfillment and violation.
MB Meta-
Its role is to select a broker that is
capable of deploying a service with
the specified user requirements.
B Broker It interacts with virtual or physical
resources, and in case the required
service needs to be deployed it in-
teracts directly with the ASD.
ASD Automatic
Service De-
It installs the required service on
the selected resource on demand
S Service The service that users want to de-
ploy and/or execute
R Resource Physical machines, on which virtual
machines can be deployed/installed
Meta Negotiator Meta Broker Broker Automatic
Service Deployer
Service Instance
Service Instance
loop (renegotiation needed)
QoS Reqs
QoS Reqs
Target broker
[QoS reqs can't be met]
Service Req
Service Req
Service Req
Service response
Deploy req
Service Req
Service response
[available service]
step a)
step b)
step c)
Figure 3: Component interactions in the SSV architecture.
The sequence diagram in Figure 3 reveals the interac-
tions of the components and the utilization steps of the
architecture, which are detailed in the following:
User starts a negotiation for executing a service with
certain QoS requirements (specified in a Service De-
scription (SD) with an SLA)
MN asks MB, if it could execute the service with the
specified requirements
MB matches the requirements to the properties of
the available brokers and replies with an acceptance
or a different offer for renegotiation
MN replies with the answer of MB. Steps 1-4 may
continue for renegotiations until both sides agree on
the terms (to be written to an SLA document)
User calls the service with the SD and SLA
MN passes SD and the possibly transformed SLA (to
the protocol the selected broker understands) to the
MB calls the selected Broker with SLA and a possi-
bly translated SD (to the language of the Broker)
The Broker executes the service with respect to the
terms of the SLA (if needed deploys the service be-
fore execution)
In the final step the result of the execution is re-
ported to the Broker, the MB, the MN, finally to
the User (or workflow engine)
Note that this utilization discussion does not reflect
cases when failures occur in the operational processes of
the components, when local adaptation is required. This
sequence represents the ideal execution flow of the SSV
architecture. In this case there is no autonomic behaviour
is needed, however in the next section we discuss and head
towards the autonomous components and their effects on
this ideal sequence (see tables 2, 3 and 4 for details).
While serving requests the architecture also processes
background tasks that are not in the ideal execution path,
these are also presented on Figure 3. The following back-
ground tasks are information collecting procedures that
provide accurate information about the current state of
the infrastructure up to the meta-brokering level:
ASD monitors the states of the virtual resources and
deployed services (step a)
ASD reports service availability and properties to its
Broker (step b)
All Brokers report available service properties to the
MB (step c)
4.2. Autonomically managed service virtualization
The previously presented SSV architecture and the de-
tailed utilization steps show that agreement negotiation,
brokering and service deployment are closely related and
each of them requires extended capabilities in order to in-
teroperate smoothly. Nevertheless each part represents an
open issue, since agreement negotiation, SLA-based ser-
vice brokering and on-demand adaptive service deploy-
ment are not supported by current solutions in diverse
distributed environments. In this subsection we focus on
illustrating how autonomic operations appear in the com-
ponents of the SSV architecture. Figure 4 shows the auto-
nomic management interfaces and connections of the three
main components: agreement negotiation, brokering and
service deployment.
We distinguish three types of interfaces in our architec-
ture: the job management interface, the negotiation inter-
Meta broker
Deployer (ASD)
Instance (S)
Manager (AM)
Job Management
Figure 4: Autonomic components in the SSV architecture.
face and the self-management interface. Negotiation inter-
faces are typically used by the monitoring processes of bro-
kers and meta-brokers during the negotiation phases of the
service deployment process. Self-management is needed
to re-negotiate established SLAs during service execution.
The negotiation interface implements negotiation proto-
cols, SLA specification languages, and security standards
as stated in the meta-negotiation document (see Figure 5).
Job management interfaces are necessary for the ma-
nipulation of services during execution, for example for
the upload of input data, or for the download of output
data, and for starting or cancelling job executions. Job
management interfaces are provided by the service infras-
tructure and are automatically utilized during the service
deployment and execution processes.
In the next section we will focus on the self-management
interface. The Autonomic manager in the SSV architec-
ture is an abstract component, that specifies how self-
management is carried out. (Later on use cases in Fig-
ure 9 will reflect, how the abstract Autonomic Manager
is realized and used in the three components of the ar-
chitecture.) All components of the architecture is notified
about the system malfunction through appropriate sensors
(see Figure 4). This interface specifies operations for sens-
ing changes of the desired state and for reacting to that
changes. Sensors can be activated using some notification
approach (e.g. implemented by the WS-Notification stan-
dard). Sensors may subscribe for specific topic, e.g. viola-
tion of the execution time. Based on the measured values
as demonstrated in [5] notifications are obtained, if exe-
cution time is violated or seems to be violated very soon.
After the activation of the control loop, i.e. propagation
of the sensed changes to the appropriate component, the
service actuator reacts and invokes proper operations, e.g.
migration of resources. An example software actuator used
for the meta-negotiations is the VieSLAF framework [5],
which bridges between the incompatible SLA templates by
executing the predefined SLA mappings. Based on various
malfunction cases, the autonomic manager propagates the
reactions to the Meta negotiatiator,Meta-broker or Au-
tomatic Service Deployer. We discuss the closed loop of
control using the the example with meta-negotiations in
the next section.
5. Required components to realize the autonomic
SSV architecture
In this section we detail three main categories, where
the basic requirements of SLA-based service virtualization
arise. We place these areas in the SSV architecture shown
in Figure 2, and detail the related parts of the proposed
solution. We also emphasize the interactions among these
components in order to build one coherent system.
In our proposed approach, users describe the require-
ments for an SLA negotiation on a high level using the con-
cept of meta-negotiations. During the meta-negotiation
only those services are selected, which understand specific
SLA document language and negotiation strategy or pro-
vide a specific security infrastructure. After the meta-
negotiation process, a meta-broker selects a broker that
is capable of deploying a service with the specified user
requirements. Thereafter, the selected broker negotiates
with virtual or physical resources using the requested SLA
document language and using the specified negotiation
strategy. Once the SLA negotiation is concluded, service
can be deployed on the selected resource using the virtu-
alization approach.
In subsection 5.1 we discuss the meta-negotiation com-
ponent and detail the autonomic negotiation bootstrap-
ping and service mediation scenario. In subsection 5.2 we
discuss the brokering functionalities of SSV and present
a self-management scenario for dealing with broker fail-
ures. Then, in subsection 5.3 we discuss autonomic ser-
vice deployment and virtualization for handling resource
and service failures. Finally, in subsection 5.4 we discuss
the autonomic capabilities of each SSV layer through rep-
resentative use cases.
5.1. Agreement negotiation
Prior Cloud infrastructures, users, who did not have
access to supercomputers or did not have local clusters
big enough, had to rely on Grid systems to execute ser-
vices requiring high performance computations. In such
environments, users had to commit themselves to dedi-
cated Grid portals to find appropriate resources for their
services, and on-demand execution was very much depen-
dent on the actual load of the appropriate Grid. Service
providers and consumers had to communicate using pro-
prietary negotiation formats supported by the particular
portal limiting the number of services a consumer may ne-
gotiate with. Nowadays, with the emergence of Clouds,
service providers and consumers need to meet each other
dynamically and on demand. Novel negotiation strategies
and formats are necessary supporting the communication
dynamics of the present day service invocations.
Before committing themselves to an SLA, the user and
the provider may enter into negotiations that determine
the definition and measurement of user QoS parameters,
and the rewards and penalties for meeting and violating
them respectively. The term negotiation strategy repre-
sents the logic used by a partner to decide which provider
or consumer satisfies his needs best. A negotiation proto-
col represents the exchange of messages during the negoti-
ation process. Recently, many researchers have proposed
different protocols and strategies for SLA negotiation in
Grids [30]. However, these not only assume that the par-
ties to the negotiation understand a common protocol but
also assume that they share a common perception about
the goods or services under negotiation. In reality how-
ever, a participant may prefer to negotiate using certain
protocols for which it has developed better strategies, over
others. Thus, the parties to a negotiation may not share
the same understanding that is assumed by the earlier pub-
lications in this field.
In order to bridge the gap between different negotia-
tion protocols and scenarios, we propose a so-called meta-
negotiation architecture [4]. Meta-negotiation is needed
by means of a meta-negotiation document where partici-
pating parties may express: the pre-requisites to be sat-
isfied for a negotiation, for example a specific authenti-
cation method required or terms they want to negotiate
on (e.g. time, price, reliability); the negotiation protocols
and document languages for the specification of SLAs that
they support; and conditions for the establishment of an
agreement, for example, a required third-party arbitrator.
These documents are published into a searchable registry
through which participants can discover suitable partners
for conducting negotiations. In our approach, the partici-
pating parties publish only the protocols and terms while
keeping negotiation strategies hidden from potential part-
The participants publishing into the registry follow a
common document structure that makes it easy to dis-
cover matching documents (as shown in Figure 5). This
document structure consists of the following main sections:
Each document is enclosed within the hmetanegotiationi
... h/meta negotiationitags. The document contains
an hentityielement defining contact information, organi-
zation and a unique ID of the participant. Each meta-
negotiation comprises three distinguishing parts, namely
pre-requisites, negotiation and agreement as described in
the following paragraph.
As shown in Figure 5, prerequisites define the role a
participating party takes in a negotiation, the security
credentials and the negotiation terms. For example, the
security element specifies the authentication and autho-
rization mechanisms that the party wants to apply before
starting the negotiation process. For example, the con-
sumer requires that the other party should be authen-
ticated through the Grid Security Infrastructure (GSI)
[12]. The negotiation terms specify QoS attributes that
a party is willing to negotiate and are specified in the
hnegotiation termielement. As an example, the ne-
gotiation terms of the consumer are beginTime, endTime,
and price. Details about the negotiation process are de-
1. <meta-negotiation ...>
2. ...
3. <pre-requisite>
4. <security>
5. <authentication value="GSI" location="uri"/>
6. </security>
7. <negotiation-terms>
8. <negotiation-term name="beginTime"/>
9. <negotiation-term name="endTime"/>
10. ...
11. </negotiation-terms>
12. </pre-requisite>
13. <negotiation>
14. <document name="WSLA" value="uri" .../>
15. <protocol name="alternateOffers"
16. schema="uri" location="uri" .../>
17. </negotiation>
18. <agreement>
19. <confirmation name="arbitrationService" value="uri"/>
20. </agreement>
Figure 5: Example Meta Negotiation Document
fined within the hnegotiationielement. Each document
language is specified within hdocumentielement. Once
the negotiation has concluded and if both parties agree
to the terms, then they have to sign an agreement. This
agreement may be verified by a third party organization
or may be lodged with another institution who will also
arbitrate in case of a dispute. Figure 6 emphasizes a meta-
negotiation infrastructure embedded into the SSV archi-
tecture as proposed in Figure 2. In the following we ex-
plain the Meta-Negotiation infrastructure.
The registry is a searchable repository for meta-negotia-
tion documents that are created by the participants. The
meta-negotiation middleware facilitates the publishing of
the meta-negotiation documents into the registry and the
integration of the meta-negotiation framework into the ex-
isting client and/or service infrastructure, including, for
example, negotiation or security clients. Besides being as
a client for publishing and querying meta-negotiation doc-
uments (steps 1 and 2 in Figure 6), the middleware delivers
necessary information for the existing negotiation clients,
i.e. information for the establishment of the negotiation
sessions (step 4, Figure 6) and information necessary to
start a negotiation (step 5 in Figure 6).
5.2. Service brokering
To deal with heterogeneity and the growing number of
services, special purpose brokers (for human-provided or
computation-intensive services) or distributed broker in-
stances should be managed together in an interoperable
way with a higher-level mediator, which is able to dis-
tribute the load of user requests among diverse infrastruc-
tures. This high-level mediator (which we call a meta-
broker) component in a unified service infrastructure needs
to monitor the states of the services and the performances
of these brokers, since brokers may have various proper-
ties, which should be expressed with a general description
language known by service consumers. These properties
may include static ones, some of which are specialized for
Figure 6: Meta-negotiation in the SSV architecture.
managing human-provided services, others for computing-
intensive or data-intensive services; and dynamic ones, e.g.
if two brokers are managing the same type of services,
some of these may have longer response times, less secure
or more reliable. Information on the available services (e.g.
type, costs, amount) also belongs to this category, since it
changes over time. The broker properties should be stored
and updated in a registry accessible by higher-level man-
agers. The update intervals of broker state changes and
the amount of data transferred should also be set auto-
matically, with respect to the number of available services
and utilized brokers. Too frequent updates could lead to
an unhealthy state and rare updates cause higher uncer-
tainty and inefficient management. Therefore this registry
should be more like a local database that makes this higher
level brokering able to decide, which broker could provide
the fittest service (according to the consumer requirements
and SLA terms).
Self-adaptability also appears at this level: generally a
bit higher uncertainty exists in broker selection compared
to service selection, since the high dynamicity of the broker
(and forwarded, filtered service) properties and availability
cause volatility in the information available at this level.
To cope with this issue, several policies could be defined
by sophisticated predicting algorithms, machine learning
techniques or random generator functions. Load balanc-
ing among the utilized brokers should also be taken into
account during broker selection. Finally basic fault tol-
erant operations, such as re-selection on broker failure or
malfunctioning also need to be handled.
In our proposed SSV architecture Brokers (B) are the
basic components that are responsible for finding the re-
quired services deployed on a specific infrastructure with
the help of the ASD. This task requires various activities,
such as service discovery, matchmaking and interactions
with information systems, service registries, repositories.
There are several brokering solutions both in Grid [22] and
SOAs[26], but agreement support is still an open issue. In
our architecture brokers need to interact with ASDs and
use adaptive mechanisms in order to fulfill the agreement.
A higher-level component is also responsible for bro-
kering in our architecture: the Meta-Broker (MB) [20].
Meta-brokering means a higher level resource management
that utilizes existing resource or service brokers to access
various resources. In a more generalized way, it acts as
a mediator between users or higher level tools (e.g. ne-
gotiators or workflow managers) and environment-specific
resource managers. The main tasks of this component
are: to gather static and dynamic broker properties (avail-
ability, performance, provided and deployable services, re-
sources, and dynamic QoS properties related to service
execution), to interact with MN to create agreements for
service calls, and to schedule these service calls to lower
level brokers, i.e. match service descriptions (SD) to bro-
ker properties (which includes broker provided services).
Finally the service call needs to be forwarded to the se-
lected broker.
Figure 7 details the Meta-Broker (MB) architecture
showing the required components to fulfill the above men-
tioned tasks. Different brokers use different service or
resource specification descriptions for understanding the
user request. These documents need to be written by
the users to specify all kinds of service-related require-
ments. In case of resource utilization in Grids, OGF [28]
has developed a resource specification language standard
called JSDL [1]. As the JSDL is general enough to describe
jobs and services of different grids and brokers, this is the
default description format of MB. The Translator com-
ponent of the Meta-Broker is responsible for translating
the resource specification defined by the user to the lan-
guage of the appropriate resource broker that MB selects
to use for a given call. These brokers have various features
for supporting different user needs, therefore an extend-
able Broker Property Description Language (BPDL) [20]
is needed to express metadata about brokers and their
offered services. The Information Collector (IC) compo-
nent of MB stores the data of the reachable brokers and
historical data of the previous submissions. This informa-
tion shows whether the chosen broker is available, or how
reliable its services are. During broker utilization the suc-
cessful submissions and failures are tracked, and regarding
these events a rank is modified for each special attribute
in the BPDL of the appropriate broker (these attributes
were listed above). In this way, the BPDL documents rep-
resent and store the dynamic states of the brokers. In
order to support load balancing, there is an IS Agent (IS
refers to Information System) reporting to IC, which reg-
ularly checks the load of the underlying resources of each
connected broker, and store this data. It also communi-
cates with the ASDs, and receives up-to-date data about
Figure 7: Meta-Broker in the SSV architecture.
the available services and predicted invocation times (that
are used in the negotiations). The matchmaking process
consists of the following steps: The MatchMaker (MM)
compares the received descriptions to the BPDL of the
registered brokers. This selection determines a group of
brokers that can provide the required service. Otherwise
the request is rejected. In the second phase the MM counts
a rank for each of the remaining brokers. This rank is
calculated from the broker properties that the IS Agent
updates regularly, and from the service completion rate
that is updated in the BPDL for each broker. When all
the ranks are counted, the list of the brokers is ordered by
these ranks. Finally the first broker of the priority list is
selected, and the Invoker component forwards the call to
the broker.
As previously mentioned, three main tasks need to be
done by MB. The first, namely the information gathering,
is done by the IS Agent, the second one is negotiation
handling and the third one is service selection. They need
the following steps: During the negotiation process the
MB interacts with MN: it receives a service request with
the service description (in JSDL) and SLA terms (in MN
document) and looks for a deployed service reachable by
some broker that is able to fulfill the specified terms. If a
service is found, the SLA will be accepted and the and MN
notified, otherwise the SLA will be rejected. If the service
requirements are matched and only the terms cannot be
fulfilled, it could continue the negotiation by modifying the
terms and wait for user approval or further modifications.
In the following we demonstrate, how principles of au-
tonomic computing are applied to this level of the SSV.
Brokers are the basic components that are responsible for
finding the required services with the help of ASD. This
task requires various activities, such as service discovery,
matchmaking and interactions with information systems
and service registries. In our architecture brokers need to
interact with ASD and use adaptive mechanisms in order
to fulfill agreements.
5.3. Service deployment and virtualization
Automatic service deployment (ASD) builds on basic
service management functionalities. It provides the dy-
namics to service based applications (SBAs) – e.g. during
the SBA‘s lifecycle services can appear and disappear (be-
cause of infrastructure fragmentation, failures, etc.) with-
out the disruption of the SBA’s overall behavior. Service
deployment process is composed of 8 steps: (i)Selection
of the node where the further steps will take place; (ii)
Installation of the service code; (iii)Configuration of the
service instance; (iv)Activation of the service to make it
public; (v)Adaptation of the service configuration while
it is active; (vi)Deactivation of the service in to sup-
port its offline modifications; (vii) Offline update of the
service code to be up-to-date, after update the new code
optionally gets reconfigured; and, (viii)Decommission of
the offline service when it is not needed on the given host
any more. Automation of deployment is automation of all
these steps.
In this article, even though there could be other ser-
vices that are built on top of service deployment, we fo-
cus and provide an overview on the various requirements
and tasks to be fulfilled by the ASD to support the meta-
negotiation and meta-brokering components of the SSV
Figure 8 shows the SSV components related to ASD
and their interconnections. The ASD is built on a repos-
itory where all deployable services are stored as virtual
appliances (VA). In this context, virtual appliance means
everything what is needed in order to deploy a service on a
selected site. The virtual appliance could be either defined
by an external entity or by the ASD’s appliance extraction
functionality [18].
To interface with the broker the ASD publishes in the
repository the deployable and the already offered services.
Thus the repository helps to define a schedule to execute a
service request taking into consideration those sites where
the service has been deployed and where it could be ex-
Cloud or Grid
Figure 8: Service deployment in SSV.
ecuted but has not yet been installed. If the requested
services are not available, the broker checks whether any
of the resources can deliver the service taking into account
the possible deployments.
According to the OGSA-Execution Planning Services
(EPS) [13] scenarios, a typical service broker has two main
connections with the outside world: the Candidate Set
Generators (CSG), and the Information Services. The
task of the CSG is to offer a list of sites, which can per-
form the requested service according to the SLA and other
requirements. Whilst the information services should of-
fer general overview about the state of the SBA, Grid or
Cloud. In most cases the candidate set generator is an in-
tegral part of the broker. Thus instead of the candidate set
adjustments, the broker queries the candidate site list as
it would do without ASD. Then the broker would evaluate
the list and as a possible result to the evaluation it would
initiate the deployment of a given service. As a result the
service call will be executed as a composed service instead
of a regular call. The composition will contain the deploy-
ment task as its starting point and the actual service call
as its dependent task. Since both the CSG and the bro-
kers heavily rely on the information system, the ASD can
influence their decision through publishing dynamic data.
This data could state service presence on sites where the
service is not even deployed.
The selected placement of the ASD depends on the site
policies on which the brokering takes place. In case the site
policy requires a strict scheduling solution then either the
CSG or the information system can be our target. If there
is no restriction then the current broker can be replaced
with an ASD extended one. In case the candidate set gen-
erator is altered then it should be a distributed, ontology-
based adaptive classifier to define a set of resources on
which the service call can be executed [36]. The CSG can
build its classification rules using the specific attributes
of the local information systems. Each CSG could have
a feedback about the performance of the schedule made
upon its candidates. The ASD extended CSG should have
three interfaces to interoperate with other CSGs and the
broker. First of all, the CSGs could form a P2P net-
work, which requires two interfaces. The first manages
the ontology of the different information systems by shar-
ing the classifier rules and the common ontology patterns
distributed as an OWL schema. The second interface sup-
ports decision-making among the peers. It enables the
forwarding of the candidate request from the broker. The
third interface lies between the broker and the CSGs to
support passing the feedback for the supervised learning
technique applied by the CSGs. This interface makes it
possible for the broker to send back a metric packet about
the success rate of the candidates.
The brokers should be adapted to ASD differently de-
pending on where the ASD extensions are placed. If both
the CSG’s and the broker’s behavior is changed then the
broker can make smarter decisions. After receiving the
candidate site set, the broker estimates the deployment
and usage costs of the given service per candidate. For
the estimation it queries the workspace service (WS – pro-
vided by the Nimbus project) that offers the virtualization
capabilities – virtual machine creation, removal and man-
agement – of a given site as a service. This service should
accept cost estimate queries with a repository entry (VA)
reference as an input. The ASD should support differ-
ent agents discovering different aspects of the deployment.
If only the broker’s behavior is changed, and the CSG
remains untouched, then the ASD would generate deploy-
ment service calls on overloaded situations (e.g. when SLA
requirements are endangered). These deployment service
calls should use the workspace service with the overloaded
service’s repository reference.
Finally it is possible to alter the information system’s
behavior. This way the ASD provides information sources
of sites, which can accept service calls after deployment.
The ASD estimates and publishes the performance related
entries (like estimated response time) of the information
system. These entries are estimated for each service and
site pair, and only those are published which are over a
predefined threshold.
Regarding component interactions, the ASD needs to
be extended with the following in order to communicate
with brokers: Requested service constraints have to be
forced independently from what Virtual Machine Monitor
(or hypervisor [2]) is used. To help the brokers making
their decisions about which site should be used the ASD
has to offer deployment cost metrics which can even be
incorporated on higher level SLAs. The ASD might initi-
ate service deployment/decommission on its own when it
can prevent service usage peaks/lows, to do so it should
be aware of the agreements made on higher levels.
In SSV there is a bidirectional connection between the
ASD and the service brokers. First the service brokers
could instruct ASD to deploy a new service (we discussed
this in detail in Section 8). However, deployments could
also occur independently from the brokers as explained in
the following. After these deployments the ASD has to
notify the corresponding service brokers about the infras-
tructure changes. This notification is required, because
information systems cache the state of the SBA for scala-
bility. Thus even though a service has just been deployed
on a new site, the broker will not direct service requests to
the new site. This is especially needed when the deploy-
ment was initiated to avoid an SLA violation.
5.4. Autonomous case studies in the SSV architecture
This subsection introduces how the self-management
and autonomous capabilities of the SSV architecture are
used in representative use cases in each SSV layer. Figure 9
summarizes the SSV components and the corresponding
self-management examples.
5.4.1. Negotiation Bootstrapping Case Study
A taxonomy of possible faults in this part of the ar-
chitecture, and the autonomic reactions to these faults are
summarized in Table 2.
Before using the service, the service consumer and the
service provider have to establish an electronic contract
defining the terms of use. Thus, they have to negotiate
the detailed terms of contract, e.g. the execution time of
the service. However, each service provides a unique nego-
tiation protocol, often expressed using different languages
representing an obstacle within the SOA architecture and
especially in emerging Cloud computing infrastructures
[6]. We propose novel concepts for automatic bootstrap-
ping between different protocols and contract formats in-
creasing the number of services a consumer may negotiate
with. Consequently, the full potential of publicly available
services could be exploited.
Figure 9 depicts how the principles of autonomic com-
puting is be applied to negotiation bootstrapping [3]. The
management is done through following steps: as a pre-
requisite of the negotiation bootstrapping users have to
specify a meta negotiation (MN) document describing the
requirements of a negotiation, as for example required ne-
gotiation protocols, required security infrastructure, pro-
vided document specification languages, etc. (More on
meta negotiation can be read in [4].) The autonomic man-
agement phases are described in the following:
Monitoring. During the monitoring phase all candi-
date services are selected, where negotiation is possible or
bootstrapping is required.
Analysis. During the analysis phase the existing knowl-
edge base is queried and potential bootstrapping strategies
are found (e.g. in order to bootstrap between WSLA and
Planning. In case of missing bootstrapping strategies
users can define new strategies in a semi-automatic way.
Execution. Finally, during the execution phase the
negotiation is started by utilizing appropriate bootstrap-
ping strategies.
5.4.2. Broker Failures Case Study
Autonomic behaviour is needed by brokers basically in
two cases: the first one is to survive failures of lower level
components, the second is to regain healthy state after lo-
cal failures. A taxonomy of the sensible failures and the
autonomic reactions of the appropriate brokering compo-
nents are gathered and shown in Table 3. The fourth case,
the broker failure, is detailed in the next paragraph. To
overcome some of these difficulties brokers in SSV use the
help of the ASD component to re-deploy some services.
In the following we present a case study showing how
autonomic principles are applied to the Meta-Broker to
handle broker failures.
Monitoring. During the monitoring phase all the in-
terconnected brokers are tracked: the IS Agent component
of the Meta-Broker gathers state information about the
brokers. The Matchmaker component also incorporates a
feedback-based solution to keep track of the performances
of the brokers.
Analysis. During the analysis phase the Information
Collector of the Meta-Broker is queried for broker avail-
ability and performance results.
Planning. In case of incoming service request the
MatchMaker component determines the ranking of the bro-
ker according to their performance data gathered in the
previous phases, and the broker with the highest rank is
selected. In case of a broker failure the ranks are recalcu-
lated and the failed broker is skipped. If the SLA cannot
be fulfilled by the new broker, SLA renegotiation is prop-
agated to MN. If no other matching broker is found a new
broker deployment request is propagated to ASD.
Execution. Finally, during the execution phase the
selected broker is invoked.
5.4.3. Self-initiated Deployment
To cope with the ever varying demand of the services
in the service based application, service instances should
form autonomous groups based on locality and on the ser-
vice interfaces they offer (e.g., neighboring instances offer-
ing the same interface should belong to the same group).
Using peer-to-peer mechanisms, the group can decide to
locally increase the processing power of a given service.
This decision could either involve to reconfigure an under-
performing service instance in the group to allow heavier
loads or it could also introduce the deployment of a new
instance to handle the increased needs.
The smallest autonomous group our architecture can
handle is formed by a single service instance. In case of
such small groups, we refer to the previously listed tech-
niques as self-initiated deployment. These single service
instances are offering self-management interfaces (seen in
Figure 4), thus they could identify erroneous situations
that could be solved by deploying an identical service in-
stance on another site. A summary of the possible erro-
neous situations and the autonomic reactions taken when
they arise is summarized in Table 4. In the following, we
Execution of a Meta
Evaluation of
Application of
existing and
definition of new
Automatic service
deployment (ASD)
SSV Component
process phase
Tracking of all
Analysis of the
results and
determins Broker
Broker with the
highest rank is
Identification of
defunct or
service state
Decision about
possible state
Deployment job
Deployment job
execution, proxy
installation for
request forwarding
Self-initiated Deployment
Figure 9: Use cases for autonomic processes in the SSV architecture
Table 2: Taxonomy of faults and autonomic reactions for Meta Negotiation.
Fault Autonomic reaction Propagation
non matching SLA templates SLA Mapping as described in [5] no, handled by the meta-negotiation layer
non matching SLA languages bootstrapping as described in [3] no, handled by the meta-negotiation layer
Table 3: Taxonomy of faults and autonomic reactions in Service brokering.
Fault Autonomic reaction Propagation
physical resource failure new service selection possible SLA renegotiation
or redeployment with ASD
service failure new service selection possible SLA renegotiation
or redeployment with ASD
wrong service response new service selection possible SLA renegotiation
broker failure new broker selection possible SLA renegotiation
or possible deployment with ASD
no service found by some broker initiate new service deployment deployment with ASD
no service found by some broker new broker selection possible SLA renegotiation
broker overloading initiate new broker deployment deployment with ASD
and possible SLA renegotiation
meta-broker overloading initiate new meta-broker deployment deployment with ASD
describe the autonomous behavior of a single service in-
Monitoring. During the monitoring phase the ASD
identifies two critical situations: (i) the autonomous ser-
vice instance becomes defunct (it is not possible to modify
the instance so it could serve requests again) and (ii) the
instance turns overloaded on such extent that the under-
lying virtual machine cannot handle more requests. To
identify these situations, service instances should offer in-
terfaces to share their health status independently from an
information service. This health status information does
not need to be externally understood, the only require-
ment that other service instances, which offer the same
interface should understand it.
Analysis. The ASD first decides whether the ser-
vice initiated deployment is required (because it was over-
loaded) or the its replication is necessary (because it has
became defunct). First in both cases the ASD identifies
the service’s virtual appliance to be deployed, then in the
latter case the ASD also prepares for state transfer of the
service before it is actually decommissioned.
Planning. After the ASD identified the service it gen-
erates a deployment job and requests its execution from
the service broker.
Table 4: Taxonomy of faults and autonomic reactions in Self-Initiated Deployment.
Fault Autonomic reaction Propagation
Degraded service
health state
Service reconfiguration
Initiate service cloning
with state transfer
If the SLA will not be violated notify service broker about change,
otherwise ask for renegotiation
Defunct service Initiate service cloning If the SLA will not be violated notify service broker about change,
otherwise ask for renegotiation
Service decommis-
Offer proxy If the proxy receives request after redirecting the request it also no-
tifies the broker about the new target to avoid later calls
Proxy lifetime ex-
Decommission service
Execution. If a service is decommissioned on the orig-
inal site, then a service proxy is placed instead on the site.
This proxy forwards the remaining service requests to the
newly deployed service using WS-Addressing. The proxy
decommissions itself when the frequency of the service re-
quests to the proxy decreases under a predefined value.
6. Evaluation of the SSV architecture in a simu-
lated environment
In order to evaluate our proposed solution, we decided
to set up a simulation environment, in which the inter-
operable infrastructure management of our approach can
be examined. For the evaluation of the performance gains
of using our proposed SSV solution, we use a typical bio-
chemical application as a case study called TINKER Con-
former Generator application [37], gridified and tested on
production Grids. The application generates conformers
by unconstrained molecular dynamics at high tempera-
ture to overcome conformational bias then finishes each
conformer by simulated annealing and energy minimiza-
tion to obtain reliable structures. Its aim is to obtain
conformation ensembles to be evaluated by multivariate
statistical modeling.
The execution of the application consists of three phases:
The first one is performed by a generator service responsi-
ble for the generation of input data for parameter studies
(PS) in the next phase. The second phase consist of a PS
sub-workflow, in which three PS services are defined for
executing three different algorithms (dynamics, minimiza-
tion and simulated annealing – we refer to these services
and the generator service as TINKERALG), and an addi-
tional PS task that collects the outputs of the three threads
and compresses them (COLL). Finally in the third phase,
a collector service gathers the output files of the PS sub-
workflows and uploads them in a single compressed file
to the remote storage (UPLOAD). These phases contain
6 services, out of which four are parameter study tasks
that are executed 50 times. Therefore the execution of the
whole workflow means 202 service calls. We set up the
simulation environment for executing a similar workflow.
For the evaluation, we have created a general simu-
lation environment, in which all stages of service execu-
tion in the SSV architecture can be simulated and coordi-
nated. We have created the simulation environment with
the help of the CloudSim toolkit [8] (that includes and ex-
tends GridSim). It supports modeling and simulation of
large scale Cloud computing infrastructure, including data
centers, service brokers and provide scheduling and allo-
cations policies. Our general simulation architecture that
builds both on GridSim and on CloudSim, can be seen in
Figure 10. On the left-bottom part we can see the GridSim
components used for the simulated Grid infrastructures,
and on the right-bottom part we can find CloudSim com-
ponents. Grid resources can be defined with different Grid
types, they consist of more machines, to which workloads
can be set, while Cloud resources are organized into Data-
centers, on which Virtual machines can be deployed. Here
service requests are called as cloudlets, which can be exe-
cuted on virtual machines. On top of this simulated Grid
and Cloud infrastructures we can set up brokers. Grid bro-
kers can be connected to one or more resources, on which
they can execute so-called gridlets (ie. service requests).
Different properties can be set to these brokers and vari-
ous scheduling policies can also be defined. A Cloud bro-
ker can be connected to a data center with one or more
virtual machines, and it is able to create and destroy vir-
tual machines during simulation, and execute cloudlets on
these virtual machines. The Simulator class is a CloudSim
entity that can generate a requested number of service re-
quests with different properties, start and run time. It is
connected to the created brokers and able to submit these
requests to them (so is acts as a user or workflow engine).
It is also connected to the realized meta-broker component
of SSV, the Grid Meta-Broker Service through its web ser-
vice interface and able to call its matchmaking service for
broker selection.
We submitted the simulated workflow in three phases:
in the first round 61 service requests for input genera-
tion, then 90 for executing various TINKER algorithms,
finally in the third round 51 calls for output preparation.
The simulation environment was set up similarly to the
real Grid environment we used for testing the TINKER
Figure 10: Simulation architecture with CloudSim.
Table 5: Deployment times of the different services in the TINKER application.
Service Average deployment time Standard deviation
GEN 8.2 sec 1.34 sec
TINKERALG 8.3 sec 1.48 sec
COLL 6.9 sec 0.84 sec
UPLOAD 6.9 sec 1.21 sec
workflow application. Estimating the real sizes of these
distributed environments, we set up four simulated Grids
(GILDA, VOCE, SEEGRID and BIOMED [10]) with 2, 4,
6 and 8 resources (each of them had 4 machines). Out of
the 202 jobs 151 had special requirements: they use the
TINKER library available in the last three Grids, which
means these calls need to be submitted to these environ-
ments, or to Cloud resources (with pre-deployed TINKER
environments). The simulated execution time of the 150
parameter study services were set to 30 minutes, the first
generator service to 90 minutes, and the other 51 were set
to 10 minutes. All of the four brokers (set to each sim-
ulated Grid one-by-one) used random resource selection
policy, and all the resources had background workload,
for which the traces were taken from the Grid Workloads
Archive (GWA) [14] (we used the GWA-T-11 LCG Grid
log file). In our simulation architecture we used 20 nodes
(called resources in the simulation), therefore we parti-
tioned the logs and created 20 workload files (out of the
possible 170 according to the number of nodes in the log).
The sorting of the job data to files from the original log file
were done continuously, and their arrival times have not
been modified, and the run time of the jobs also remained
the same. According to these workload files the load of the
simulated environments are shown in Figure 11 (which are
also similar to the load experienced on the real Grids).
One Cloud broker has also been set up. It managed four
virtual machines deployed on a data center with four hosts
of dual-core CPUs. In each simulation all the jobs were
sent to the Meta-Broker to select an available broker for
submission. It takes into account the actual background
load and the previous performance results of the brokers
for selection. If the selected Grid broker had a background
load that exceeded a predefined threshold value, it selected
the Cloud broker instead.
Out of the 202 workflow services 151 use TINKER bi-
naries (three different algorithms are executed 50 times
plus one generator job). These requirements can be for-
mulated in SLA terms, therefore each service of the work-
flow has an SLA request. If one of these requests are sent
to the Cloud broker, it has to check if a virtual machine
(VM) has already been created that is able to fulfil this
request. If there is no such VM, it deploys one on-the-fly.
For the evaluation we used three different Cloud broker
configurations: in the first one four pre-deployed VMs are
used – one for each TINKER algorithm (TINKERALG)
and one for data collecting (capable for both COLL and
UPLOAD, used by the last 51 jobs). In the second case we
used only one pre-deployed VM, and deployed the rest on-
the-fly, when the first call arrived with an SLA. Finally in
the third case, when a VM received more then 15 requests
the Cloud broker duplicated it (in order to minimize the
overall execution time).
Regarding on-demand deployment, we have created 4
virtual appliances encapsulating the four different services
our TINKER workflow is based on (namely TINKERALG,
COLL and UPLOAD – we defined them in the beginning
of this section). Then we have reduced the size of the cre-
Figure 11: Workload of simulated Grids.
ated appliances with ASD’s virtual appliance optimization
facility. Finally we have deployed each service 50 times on
an 8 node (32 CPU) Eucalyptus [27] cluster, and measured
the interval between the deployment request and the ser-
vice’s first availability. Table 5 shows the measurement
results for the TINKERALG, COLL and UPLOAD im-
ages. These latencies were also applied in the simulation
environment within the Cloud broker.
In order to evaluate the performance of our proposed
SSV solution we compare it to a general meta-brokering
architecture used in Grid environments. Using this ap-
proach we created four simulations: in the first one we use
only grid brokers by the Meta-Broker (denoted by MB in
the figures) to reach grid resources of the simulated Grids.
In the second, third and fourth case we extend the match-
making of the Meta-Broker (in order to simulate the whole
SSV): when the background load of the selected grid bro-
ker exceeds 113%, it selects the Cloud broker instead to
perform Cloud bursting. In the second case the Cloud bro-
ker has four pre-deployed VMs (4VMs), while in the third
case only one, and later creates three more as described be-
fore (1+3VMs), and in the fourth it has one pre-deployed
and creates at most 7 more on demand (1+3+4VMs).
In Figure 12 we can see detailed execution times for
the simulations using purely Grid resources and the first
strategy using 4 VMs for Cloud bursting. The first phase
of the workflow has been executed only on Grid resources
in both strategies. Around the 60th job submission the
load of Grid resources exceeded the predefined overload
threshold, therefore the Cloud Broker has been selected by
the GMBS to move further job submissions to Cloud re-
sources. The arrows in the figure denote the places, where
Cloud VMs managed to keep constant load by taking over
jobs from the Grids. In Figure 13 we denoted the second
strategy, where initially only 1 VM was made available
for Cloud bursting. The rest 3 VMs have been started
on-demand, after the first VM got overloaded. In this
Figure 12: Detailed run times for Cloud bursting with 4 VMs.
Figure 13: Detailed run times for Cloud bursting with 1+3 VMs.
case we could save some operational costs for 3 VMs for
a short period, but later delayed startup times of the 3
VMs resulted in longer run times for the last 60 jobs of
the workflow. Finally, in Figure 14 we can see the third
strategy, in which we performed service duplication with
4 additional VMs. In this case we also used on-demand
VM startups, but the doubled service instances managed
to keep the overall Cloud load down.
In Figure 15 we can see the evaluation results denot-
ing the average aggregated execution times of the service
requests. From these results we can clearly see that the
simulated SSV architecture overperforms the former meta-
brokering solution using only Grid resources. Comparing
the different deployment strategies we can see that on de-
mand deployment introduces some overhead (4VMs was
faster then 1+3VMs), but service duplication (1+3+4VMs)
can enhance the performance and help to avoid SLA vio-
lations with additional VM deployment costs.
We have to mention that the SSV architecture puts
Figure 14: Detailed run times for Cloud bursting with 1+3+4 VMs.
Figure 15: Average request run times.
some overhead on service executions. Regarding the over-
head the meta-brokering layer generates, we only need to
consider the latency of the web service calls and the match-
making time of the GMBS. In this evaluation, this latency
took up around 200 milliseconds for a call (which is less
than 0.1% for all the calls of the application compared to
the total execution time we measured in the simulation),
and it is also negligible comparing to general brokering re-
sponse times (which can last up to several minutes in grid
environments). At the cloud brokering layer, VM startup
times also affect the overall execution times, as we have
seen in Figure 13. In our opinion, these latencies are ac-
ceptable, hence our automated service execution can man-
age a federation of heterogeneous infrastructure environ-
ments by providing SLA-based service executions under
such conditions, where a manual user interaction with a
single infrastructure would fail.
7. Conclusions
In a unified system consisting of heterogeneous, dis-
tributed service-based environments such as Grids, SBAs
and Clouds, there is an emerging need for transparent,
business-oriented autonomic service execution. In the fu-
ture more and more companies will face the problem of un-
foreseen, occasional demand for a high number of comput-
ing resources. In this paper we investigated how such prob-
lems could arise and gathered the requirements for a ser-
vice architecture that is able to cope with these demands.
We addressed these requirements to develop a functionally
and architecturally well-designed solution, and proposed a
novel approach called Service-level agreement-based Ser-
vice Virtualization (SSV). The presented general archi-
tecture is built on three main components: the Meta-
Negotiatior responsible for agreement negotiations, the Meta-
Broker for selecting the proper execution environment, and
the Automatic Service Deployer for service virtualization
and on-demand deployment.
We have also discussed how the principles of auto-
nomic computing are incorporated to the SSV architecture
to cope with the error-prone virtualization environments,
and demonstrated how autonomic actions are triggered re-
sponding to various possible failures in the service infras-
tructure. The autonomic reactions are demonstrated with
corresponding case studies. The operation of the proposed
architecture is exemplified through highlighting the cases
where the autonomous properties of the components of the
architecture were activated in order to cope with failures
during service executions.
Finally the SSV architecture is validated in a simula-
tion environment based on CloudSim, using a general bio-
chemical application as a case study. The evaluation re-
sults clearly fulfil the expected utilization gains compared
to a less heterogeneous Grid solution.
8. Acknowledgment
The research leading to these results has received fund-
ing from the European Community’s Seventh Framework
Programme FP7/2007-2013 under grant agreement 215483
[1] A. Anjomshoaa, F. Brisard, M. Drescher, D. Fellows,
A. Ly, S. McGough, D. Pulsipher, and A. Savva.
Job submission description language (JSDL) spec-
ification, version 1.0. Technical report, 2005.
[2] P. Barham, B. Dragovic, K. Fraser, S. Hand, T. Harris, A. Ho,
R. Neugebauer, I. Pratt, and A. Warfield. Xen and the art
of virtualization. In SOSP ’03: Proceedings of the nineteenth
ACM symposium on Operating systems principles, pages 164–
177, New York, NY, USA, 2003. ACM.
[3] I. Brandic, D. Music, S. Dustdar. Service Mediation and Ne-
gotiation Bootstrapping as First Achievements Towards Self-
adaptable Grid and Cloud Services. In Proceedings of Grids
meet Autonomic Computing Workshop. ACM. June, 2009.
[4] I. Brandic, D. Music, S. Dustdar, S. Venugopal, and R. Buyya.
Advanced qos methods for grid workflows based on meta-
negotiations and sla-mappings. In The 3rd Workshop on Work-
flows in Support of Large-Scale Science, pages 1–10, November
[5] I. Brandic, D. Music, P. Leitner, S. Dustdar. VieSLAF Frame-
work: Enabling Adaptive and Versatile SLA-Management. In
the 6th International Workshop on Grid Economics and Busi-
ness Models 2009 (Gecon09), 2009.
[6] R. Buyya, C. S. Yeo, S. Venugopal, J. Broberg, and I. Brandic.
Cloud computing and emerging it platforms: Vision, hype, and
reality for delivering computing as the 5th utility. Future Gen-
eration Computer Systems, 2009.
[7] R. Buyya, R. Ranjan, and R. N. Calheiros. InterCloud: Utility-
Oriented Federation of Cloud Computing Environments for
Scaling of Application Services. Lecture Notes in Computer
Science: Algorithms and Architectures for Parallel Processing.
Volume 6081, 2010.
[8] R. Buyya, R. Ranjan and R. N. Calheiros. Modeling and Sim-
ulation of Scalable Cloud Computing Environments and the
CloudSim Toolkit: Challenges and Opportunities. In proc. of
the 7th High Performance Computing and Simulation Confer-
ence, 2009.
[9] M. Q. Dang and J. Altmann. Resource allocation algorithm for
light communication grid-based workflows within an sla context.
Int. J. Parallel Emerg. Distrib. Syst., 24(1):31–48, 2009.
[10] Enabling Grids for E-sciencE (EGEE) project website., 2008.
[11] A.J. Ferrer et. al. OPTIMIS: a Holistic Approach to Cloud
Service Provisioning. Future Generation Computer Systems,
vol. 28, pp. 66–77, 2012.
[12] I. Foster, C. Kesselman, G. Tsudik, and S. Tuecke. A security
architecture for computational grids. In CCS ’98: Proceedings
of the 5th ACM conference on Computer and communications
security, pages 83–92, New York, NY, USA, 1998. ACM.
[13] I. Foster, H. Kishimoto, A. Savva, D. Berry, A. Djaoui,
A. Grimshaw, B. Horn, F. Maciel, F. Siebenlist, R. Sub-
ramaniam, J. Treadwell, and J. Reich. The open grid
services architecture, version 1.5. Technical report, 2006.
[14] The Grid Workloads Archive website.,
[15] R. Howard and L. Kerschberg. A knowledge-based framework
for dynamic semantic web services brokering and management.
In DEXA ’04: Proceedings of the Database and Expert Sys-
tems Applications, 15th International Workshop, pages 174–
178, Washington, DC, USA, 2004. IEEE Computer Society.
[16] A. Iosup, T. Tannenbaum, M. Farrellee, D. Epema, and
M. Livny. Inter-operating grids through delegated matchmak-
ing. Sci. Program., 16(2-3):233–253, 2008.
[17] K. Keahey, I. Foster, T. Freeman, and X. Zhang. Virtual
workspaces: Achieving quality of service and quality of life in
the grid. Sci. Program., 13(4):265–275, 2005.
[18] G. Kecskemeti, G. Terstyanszky, P. Kacsuk, and Zs. Nemeth.
An Approach for Virtual Appliance Distribution for Service De-
ployment. Future Generation Computer Systems, vol. 27, issue
3, pp 280–289, 2011.
[19] J.O. Kephart, D.M. Chess. The vision of autonomic computing.
Computer . 36:(1) pp. 41-50, Jan 2003.
[20] A. Kertesz and P. Kacsuk. GMBS: A New Middleware Service
for Making Grids Interoperable. In Future Generation Com-
puter Systems, vol. 26, no. 4, pp. 542-553, 2010.
[21] C. Kesselman and I. Foster. The Grid: Blueprint for a
New Computing Infrastructure. Morgan Kaufmann Publishers,
November 1998.
[22] K. Krauter, R. Buyya, and M. Maheswaran. A taxonomy and
survey of grid resource management systems for distributed
computing. Software: Practice and Experience, 32(2):135–164,
[23] I. Krsul, A. Ganguly, J. Zhang, J. A. B. Fortes, and R. J.
Figueiredo. Vmplants: Providing and managing virtual ma-
chine execution environments for grid computing. In SC ’04:
Proceedings of the 2004 ACM/IEEE conference on Supercom-
puting, Washington, DC, USA, 2004. IEEE Computer Society.
[24] K. Leal, E. Huedo, I. M. Llorente. A decentralized model for
scheduling independent tasks in Federated Grids. Future Gen-
eration Computer Systems, Volume 25, Issue 8, pp. 840–852,
[25] K. Lee, N.W. Paton, R. Sakellariou, E. Deelman, A.A.A. Fer-
nandes, G. Mehta. Adaptive Workflow Processing and Execu-
tion in Pegasus. In Proceedings of 3rd International Workshop
on Workflow Management and Applications in Grid Environ-
ments, pages 99–106, 2008.
[26] Y. Liu, A. H. Ngu, and L. Z. Zeng. Qos computation and polic-
ing in dynamic web service selection. In WWW Alt. ’04: Pro-
ceedings of the 13th international World Wide Web conference
on Alternate track papers & posters, pages 66–73, New York,
NY, USA, 2004.
[27] D. Nurmi, R. Wolski, C. Grzegorczyk, G. Obertelli, S. Soman,
L. Youseff and D. Zagorodnov. The eucalyptus open-source
cloud-computing system. In CCGRID, pages 124–131. IEEE
Computer Society, 2009.
[28] The Open Grid Forum website., 2010.
[29] D. Ouelhadj, J. Garibaldi, J. MacLaren, R. Sakellariou, and
K. Krishnakumar. A multi-agent infrastructure and a service
level agreement negotiation protocol for robust scheduling in
grid computing. In Proceedings of the 2005 European Grid
Computing Conference (EGC 2005), February 2005.
[30] M. Parkin, D. Kuo, J. Brooke, and A. MacCulloch. Challenges
in eu grid contracts. In Proceedings of the 4th eChallenges Con-
ference, pages 67–75, 2006.
[31] D. M. Quan and J. Altmann. Mapping a group of jobs in the
error recovery of the grid-based workflow within sla context.
Advanced Information Networking and Applications, Interna-
tional Conference on, 0:986–993, 2007.
[32] D. Reed, I. Pratt, P. Menage, S. Early, and N. Stratford.
Xenoservers: Accountable execution of untrusted programs. In
In Workshop on Hot Topics in Operating Systems, pages 136–
141, 1999.
[33] B. Rochwerger, D. Breitgand, E. Levy, A. Galis, K. Nagin. I.
Lloriente, R. Montero, Y. Wolfsthal, E. Elmroth, J. Caceres, M.
Ben-Yehuda W. Emmerich, F. Galan. The RESERVOIR Model
and Architecture for Open Federated Cloud Computing. IBM
Journal of Research and Development, 53(4), 2009.
[34] I. Rodero, F. Guim, J. Corbalan, L. Fong, Y. Liu, and S. Sad-
jadi. Looking for an evolution of grid scheduling: Meta-
brokering. In Grid Midd leware and Services Challenges and
Solutions, pages 105–119. Springer US, 2008.
[35] M. Surridge, S. Taylor, D. De Roure, and E. Zaluska. Expe-
riences with gria – industrial applications on a web services
grid. In E-SCIENCE ’05: Proceedings of the First International
Conference on e-Science and Grid Computing, pages 98–105,
Washington, DC, USA, 2005. IEEE Computer Society.
[36] M. Taylor, C. Matuszek, B. Klimt, and M. Witbrock. Au-
tonomous classification of knowledge into an ontology. In The
20th International FLAIRS Conference (FLAIRS), 2007.
[37] TINKER Conformer Generator workflow.,
[38] H. N. Van, F. D. Tran, and J. Menaud. Autonomic virtual
resource management for service hosting platforms. In Proceed-
ings of the ICSE Workshop on Software Engineering Challenges
of Cloud Computing, pages 1–8, 2009.
[39] C. Vazquez, E. Huedo, R. S. Montero et I. M. Llorente. Fed-
eration of TeraGrid, EGEE and OSG Infrastructures through a
Metascheduler. Future Generation Computer Systems 26, pp.
979–985, 2010.
[40] S. Venugopal, R. Buyya, and L. Winton. A grid service
broker for scheduling e-science applications on global data
grids. Concurrency and Computation: Practice and Experi-
ence, 18(6):685–699, 2006.
[41] C. A. Yfoulis, and A. Gounaris. Honoring SLAs on cloud com-
puting services: a control perspective. In Proceedings of the
European Control Conference, 2009.
... However, how the application and the infrastructure become the elements which align with the environment characteristics, such as the alignment view in our model, is undefined. Meanwhile, Kertesz, (2014) focuses on the requirements for the adaptation ability through the model of the SLA-based service visualization. In this research, how the SLA is defined by considering its appropriateness with the needs of the organization is not discussed as in our proposed model. ...
... Meanwhile, the non-functional goal used as a quality reference for the delivery of cloud service is represented in the form of the service level agreement (SLA). However, according to Kertesz, (2014), the SLA for the environment of cloud computing will involve services in a complex business process, which can hinder the achievement of the SLA. This can be caused by system failure, component changes, work load and external conditions, and hardware and software failure so that the interaction of users and the system during the SLA negotiation and service execution can fail. ...
Full-text available
The easy access and flexibility provided by cloud computing have made various types of organizations use this technology as an alternative solution. However, the decision to use the cloud technology sometimes ignores the basic aspects related to the understanding of its compliance with the characteristics of the organization. Thus, many advantages of the adoption of the technology do not align with the needs. This paper discusses the requirements engineering of the alignment of the organizational characteristics and the cloud computing technology in order to create the adaptive ability to guide strategic-business driven organizations. The model proposed in this article is formulated based on three views: architectural, alignment, and adaptive, by employing the Goal-Oriented Requirements Engineering. As a case study, the model is applied to cloud computing for university. The architectural view can translate the environmental characteristics of organizations through goal decomposition into detailed needs. Meanwhile, the alignment view can meet the implementation, so the dynamic adaptive ability can be realized through the adaptivity view.The mechanism of our proposed model can cover the adaptation needs before the migration (pre-migration), during the migration, and after the migration (post-migration).
... Small scale service providers from different geographic areas interconnect their technical infrastructure to seamless and on demand service provisioning. Interconnected distributed resources in cloud paradigm provide more elasticity during flash-crowd to maintain the consistent service level agreement (SLA) (Ouardi et al., 2017;Kertész et al., 2014). In overwhelming condition of resources state, users get serviced from optimal one transparently amongst interconnected resource domains. ...
... Small scale service providers from different geographic areas interconnect their technical infrastructure to seamless and on demand service provisioning. Interconnected distributed resources in cloud paradigm provide more elasticity during flash-crowd to maintain the consistent service level agreement (SLA) (Ouardi et al., 2017;Kertész et al., 2014). In overwhelming condition of resources state, users get serviced from optimal one transparently amongst interconnected resource domains. ...
... Customers are allowed to utilise the computing resources presented by multiple service providers [6][7][8]. These services reveal the foundations which are not only adopted but also installed, progressed or depicted virtualisation [9,10]. It is a method of transmission of services that its authenticity, scalability, and accessibility are high [11][12][13]. ...
Full-text available
Cloud computing platforms propose applied directional information technology services to its clients. It provides services according to demands by which shared resources, data, software, and other tools are to be made available based on the client's request. Load balancing is considered as one of the main issues in cloud computing. The quality of service (QoS) factors such as resources application, answering time, process time, scalability, throughput, the durability of the system, and energy consumption are improved by the balanced load. The cloud tasks scheduling and load balancing are considered as an NP-hard issue. Therefore, in this study, a new energy-aware cloud load balancing method is proposed by usage of preference alignments and ant colony optimisation algorithm. This approach aims at balancing the virtual machines workload and energy consumption of cloud data centres and also assures QoS for users. CloudSim simulator is used to confirm the performance of the method. It shows that the proposed method reduces waiting time and energy consumption on a virtual machine. Nomenclature T i ith subtask of DAG P k kth processor of the system T collection of a subtask of DAG P collection of processors in a computing environment N number of subtasks of DAG E number of the edge of DAG M number of processors P max maximum energy to be used while the server is completely in use K part of energy utilised by the idle server (i.e. 70%) U CPU utilisation T entry input subtask T exit output subtask Succ T i collection of immediate successors of T i
... However, this research lack on Virtual Machine Scheduling optimization that we have enhanced in our paper. A. Kertesz [5] propose a novel holistic architecture called SLA-based Service Virtualization (SSV) made on agreement negotiation, brokering and service deployment combined with business-oriented operation. They also use the Autonomic Computing technology to inspect their introduced architecture and study how the principles of Autonomic Computing emerge in the basic components of the architecture in order to face with changing user requirements and on-demand failure handling. ...
Full-text available
p class="0abstract">Cloud Computing conducts application, infrastructure services or platform to a very large amount of users with more choices and continuous changing requirements. Cloud providers are occupied in organizing data warehouses to arrange the continuous growth in cloud user’s acceptance. Features of cloud computing services have afforded an important tendency of companies choosing these services. In this case, many cloud users, who intend a certain service, and many cloud providers, who provision those services, create a competitive market. When constructing a Cloud scheduling scheme, the project trade-offs of the Cloud architecture should be rated. The simplest manner to rate this infrastructure is to use a simulation tool called CloudSim simulation toolkit. To reduce/avoid SLA violations in the cloud computing system, in our work we propose a SLA reduction framework in which we have considered three steps: Migration control of VMs, Energy efficiency and VmScheduling. It is also implemented in Cloudsimulator.</p
... For instance, self-manageable cloud services are discussed in [Bra09]. In [KKB14], autonomic computing principles are applied to avoid service-level agreement (SLA) violations. The authors of [TBB+15;TBB+17] propose an architecture for self-managing cloud-native applications. ...
High Performance Computing (HPC) enables significant progress in both science and industry. Whereas traditionally parallel applications have been developed to address the grand challenges in science, as of today, they are also heavily used to speed up the time-to-result in the context of product design, production planning, financial risk management, medical diagnosis, as well as research and development efforts. However, purchasing and operating HPC clusters to run these applications requires huge capital expenditures as well as operational knowledge and thus is reserved to large organizations that benefit from economies of scale. More recently, the cloud evolved into an alternative execution environment for parallel applications, which comes with novel characteristics such as on-demand access to compute resources, pay-per-use, and elasticity. Whereas the cloud has been mainly used to operate interactive multi-tier applications, HPC users are also interested in the benefits offered. These include full control of the resource configuration based on virtualization, fast setup times by using on-demand accessible compute resources, and eliminated upfront capital expenditures due to the pay-per-use billing model. Additionally, elasticity allows compute resources to be provisioned and decommissioned at runtime, which allows fine-grained control of an application's performance in terms of its execution time and efficiency as well as the related monetary costs of the computation. Whereas HPC-optimized cloud environments have been introduced by cloud providers such as Amazon Web Services (AWS) and Microsoft Azure, existing parallel architectures are not designed to make use of elasticity. This thesis addresses several challenges in the emergent field of High Performance Cloud Computing. In particular, the presented contributions focus on the novel opportunities and challenges related to elasticity. First, the principles of elastic parallel systems as well as related design considerations are discussed in detail. On this basis, two exemplary elastic parallel system architectures are presented, each of which includes (1) an elasticity controller that controls the number of processing units based on user-defined goals, (2) a cloud-aware parallel execution model that handles coordination and synchronization requirements in an automated manner, and (3) a programming abstraction to ease the implementation of elastic parallel applications. To automate application delivery and deployment, novel approaches are presented that generate the required deployment artifacts from developer-provided source code in an automated manner while considering application-specific non-functional requirements. Throughout this thesis, a broad spectrum of design decisions related to the construction of elastic parallel system architectures is discussed, including proactive and reactive elasticity control mechanisms as well as cloud-based parallel processing with virtual machines (Infrastructure as a Service) and functions (Function as a Service). To evaluate these contributions, extensive experimental evaluations are presented.
... Huedo et al. [26] used an SLA-based service virtualization which a self-manageable architecture gives an approach for easing interoperable service executions in an assorted, heterogeneous, distributed and virtualized universe of cloud services. In a distributed environment, to achieve reliable and efficient service operation, a model with the combination of negotiation, brokering and deployment with SLA-aware extensions and autonomic computing principles was used. ...
Full-text available
Cloud interoperability provides cloud services such as Software as a Service (SaaS) or customer system to communicate between the cloud providers. However, one of the most important barriers for existing researches was to adopt the application’s or data’s in cloud computing environments so as to obtain efficient cloud interoperability. This paper focuses on reliable cloud interoperability with a heterogeneous cloud computing resource environment with the objective of providing unilateral provision computing capabilities of a cloud server without the help of human interaction and allowing proper utilization of applications and services across various domains by using an effective cloud environment available at runtime. Moreover, the framework uses hybrid squirrel search genetic algorithm (HSSGA) to select the relevant features from a set of extracted features in order to eliminate irrelevant data which provides advantages of low computational time and less memory usage. Thereafter, for a proper selection of cloud server with respect to the selected features, the system has developed the improved adaptive neuro-fuzzy inference system (I-ANFIS) which provides accurate server selection and helps against uncertainties caused by servers or applications. Hence, the experimental result of the proposed framework gives an accuracy of 94.24% and remains more efficient compared to existing frameworks.
... We considered the relevance of these articles on the basis of our subject and of them chose finally 33 articles (Table 4) and included them in our review. (Group & Committee, 2006) 13 (Jrad, Tao, Brandic, & Streit, 2015) 14 (Karadsheh, 2012) 15 (Kertesz, Kecskemeti, & Brandic, 2014) 16 (Lamanna, Skene, & Emmerich, 2003) 17 (Larson, 1998) 18 (Liu & Ma, 2005) (Nikolow, et al., 2013). ...
Conference Paper
Full-text available
The economic governance in the Eurozone and the EU is not the same after the crisis. The results of the financial crisis changed the EU’s way of thinking about addressing issues such as the public debt service, the banking system functioning, the ECB’s role and the depth of its intervention. When R. Mundell (1961) developed the theory of optimum currency areas the Eurozone was certainly not in his mind, while the same view is maintained by P. Krugman (2009) too. However, the EU financial and economic sector is entirely different today. After tens of EU bodies meetings there is now a number of new rescue mechanisms, institutions and regulations. But, are all these changes well enough to get the Eurozone out of the crisis? Certainly not, but they had never been given in the past as well. The “moral hazard” and the lack of a last resort lender did not finally work as the EU leaders thought. The Eurozone is the core of the EU and it is slowly evolved, step by step, from an economic to a political union. The EU has at its disposal the European Stability Mechanism (ESM) with a lending capacity of over half a billion euros, while there have been also other mechanisms like the EFSF and the EFSM, with the total sum of granted loans by these three mechanisms to date being amounted to over 350 billion euros. Additionally, the EU heads rapidly for a banking union through the establishment of new institutions like the Single Resolution Fund (SRF), the Single Resolution Mechanism (SRM) and the Single Supervisory Mechanism (SSM), which will supervise over 180 systemic banks in the Eurozone. Furthermore, the ECB used non-standard monetary policy mechanisms in order to support countries and banking institutions. The best known mechanism is the Quantitative Easing (QE) programme worth about one trillion euros. The new model of economic governance moves to a progressive direction, but the size of the instruments may not be enough for the EU to exit the crisis. Pages 29 - 45
... For the NIST, a cloud broker may include service intermediation, aggregation/federation and arbitrage [21]. Such tasks require modelling resources (single and federated), businesses and their relationships as well as methodologies for partner discovery, resource negotiation and resource provision [1,30,25,6,28,37,8,27,10,35,15,19,23,18]. Habib, Ries and Mühlhäuser (2010) identify several parameters to select reliable providers [14]: (i ) SLA templates; (ii ) certification; (iii) portability; (iv ) interoperability; (v ) location of the data centre; (vi ) customer support facilities; (vii ) performance tests; (viii) deployment models; (ix ) federated identity management; (x ) security measures; and (xi ) user recommendation, feedback and publicity [1]. ...
Full-text available
This paper analyses the impact of trust and reputation modelling on CloudAnchor, a business-to-business brokerage platform for the transaction of single and federated resources on behalf of Small and Medium Sized Enterprises (SME). In CloudAnchor, businesses act as providers or consumers of Infrastructure as a Service (IaaS) resources. The platform adopts a multi-layered multi-agent architecture, where providers, consumers and virtual providers, representing provider coalitions, engage in trust & reputation-based provider look-up, invitation, acceptance and resource negotiations. The goal of this work is to assess the relevance of the distributed trust model and centralised fuzzified reputation service in the number of resources successfully transacted, the global turnover, brokerage fees, losses, expenses and time response. The results show that trust and reputation based brokerage has a positive impact on the CloudAnchor performance by reducing losses and the execution time for the provision of both single and federated resources and increasing considerably the number of federated resources provided.
Full-text available
p> A key component of ensuring that services are available on the cloud at the right time in the right manner is adequate cloud management. This makes it possible to provide services that meets user demands. The purpose of this research is to carry out a systematic study of management and monitoring on the cloud. Three facets were applied in conducting the categorization. These are the contribution, research, and topic facets. The purpose was to determine the level of work so far carried out in the field of cloud management. This enabled the creation of a pictorial representation of the research coverage. The result of the study showed that there are no opinion research on cloud management. Generally, articles on experience research, philosophical research and metric are the lowest at 6.62%, 4.41% and 1.90% respectively, while articles on models, solution research and evaluation research are the highest with 52.38%, 46.32% and 31.62% respectively. The outcome of this study will stimulate further research in the area cloud management and systematic studies. </p
Full-text available
The grid vision of a single computing utility has yet to materialize: while many grids with thousands of processors each exist, most work in isolation. An important obstacle for the effective and efficient inter-operation of grids is the problem of resource selection. In this paper we propose a solution to this problem that combines the hierarchical and decentralized approaches for interconnecting grids. In our solution, a hierarchy of grid sites is augmented with peer-to-peer connections between sites under the same administrative control. To operate this architecture, we employ the key concept of delegated matchmaking, which temporarily binds resources from remote sites to the local environment. With trace-based simulations we evaluate our solution under various infrastructural and load conditions, and we show that it outperforms other approaches to inter-operating grids. Specifically, we show that delegated matchmaking achieves up to 60% more goodput and completes 26% more jobs than its best alternative.
Full-text available
Successful realization of the Open Grid Services Architecture (OGSA) vision of a broadly applicable and adopted framework for distributed system integration, virtualization, and management requires the definition of a core set of interfaces, behaviors, resource models, and bindings. This document, produced by the OGSA working group within the Global Grid Forum (GGF), provides a high-level OGSA definition. The document focuses on requirements and the scope of important capabilities required to support Grid systems and applications in both e-science and e-business. The capabilities described are Execution Management, Data, Resource Management, Security, Self-Management, and Information. The description of each capability includes, to some extent, possible interrelationships with other capabilities. Capabilities are, however, largely independent of each other and there is no requirement that they all be present in an OGSA system. Further, capabilities themselves are not monolithic and it is possible that only a subset of a given OGSA capability may be present in an implementation of an OGSA system.
Full-text available
This document specifies the semantics and structure of the Job Submission Description Language (JSDL). JSDL is used to describe the requirements of computational jobs for submission to resources, particularly in Grid environments, though not restricted to the latter. The document includes the normative XML Schema for the JSDL, along with examples of JSDL documents based on this schema.
Full-text available
This work contains a short survey of recent results in the literature with a view to opening up new research direc-tions for the problem of honoring SLAs on cloud computing services. This is a new problem that has attracted significant interest recently, due to the urgent need for providers to provide reliable, customized and QoS guaranteed computing dynamic environments for end-users as agreed in contracts on the basis of certain Service Level Agreements (SLAs). Honoring SLAs is a multi-faceted problem that may involve optimal use of the available resources, optimization of the system's performance and availability or maximization of the provider's revenue and it poses a significant challenge for researchers and system administrators due to the volatile, huge and unpredictable Web environments where these computing systems reside. The use of algorithms possessing run-time adaptation features, such as dynamic resource allocation, admission control and optimiza-tion becomes an absolute must. As a continuation of the recent successful application of control theory concepts and methods to the computing systems area, our survey indicates that the problem of honoring SLAs on cloud computing services is a new interesting application for control theory and that researchers can benefit significantly from a number of well-known modern control methodologies, such as hybrid, supervisory, hierarchical and model predictive control.
Conference Paper
Nowadays, novel computing paradigms as for example Grid or Cloud Computing are gaining more and more on importance. In case of Cloud Computing users pay for the usage of the computing power provided as a service. Beforehand they can negotiate specific functional and non-functional requirements relevant for the application execution. However, providing computing power as a service bears different research challenges. On the one hand dynamic, versatile, and adaptable services are required, which can cope with system failures and environmental changes. On the other hand, human interaction with the system should be minimized. In this paper we present the first results in establishing adaptable, versatile, and dynamic services considering negotiation bootstrapping and service mediation achieved in context of the Foundations of Self-Governing ICT Infrastructures (FoSII) project. We discuss novel meta-negotiation and SLA mapping solutions for Grid/Cloud services bridging the gap between current QoS models and Grid/Cloud middleware and representing important prerequisites for the establishment of autonomic Grid/Cloud services. We present document models for the specification of meta-negotiations and SLA mappings. Thereafter, we discuss the sample architecture for the management of meta-negotiations and SLA mappings.
The various ways of implementation of virtual workspaces and their use to build layered deployment environments in the Grid, are described. A virtual workspace is an abstraction of an execution environment that can be made dynamically available to authorized clients by using well-defined protocols. Virtual machines allow a client to create a custom execution environment configured with an operating system, software stack, and access policies and deploy it on any resource running a hypervisor. A workspace description should contain sufficient information for a deployment service to create the environment represented by this workspace. Workspaces make it easier for resource owners to contribute resources to the Grid and to manage virtual organizations.
The resource management system is the central component of network computing systems. There have been many projects focused on network computing that have designed and implemented resource management systems with a variety of architectures and services. In this paper, we develop a comprehensive taxonomy for describing resource management architectures. We use this taxonomy in identifying approaches followed in the implementation of real resource management systems for large-scale network computing systems known as Grids. We use the taxonomy and the survey results to identify architectural approaches that have not been fully explored in the research.
Conference Paper
The grid vision of a single computing utility has yet to materíalize: while many grids with thousands of processors each exist, most work in isolation. An important obstacle for the effective and efficient inter-operation of grids is the problem of resource selection. In this paper we propose a solution to this problem that combines the hierarchical and decentralized approaches for interconnecting grids. In our solution, a hierarchy of grid sites is augmented with peer-to-peer connections between sites under the same administrative control. To operate this architecture, we employ the key concept of delegated matchmaking, which temporarily binds resources from remote sites to the local environment. With trace-based simulations we evaluate our solution under various infrastructural and load conditions, and we show that it outperforms other approaches to inter-operating grids. Specifically, we show that delegated matchmaking achieves up to 60% more goodput and completes 26% more jobs than its best alternative.