Content uploaded by Dominic P.A. Greenwood
Author content
All content in this area was uploaded by Dominic P.A. Greenwood
Content may be subject to copyright.
Autonomic Service Access Management for Next
Generation Converged Networks
Monique Calisti, Roberto Ghizzioli, Dominic Greenwood
Abstract. This chapter presents the Living Systems Autonomic Service Ac-
cess Management Suite, LS/ASAM, a comprehensive middleware solution en-
abling adaptive connectivity management of nomadic end hosts across hetero-
geneous access networks with autonomic optimization of network performance
and availability.
1. Introduction
Next Generation Networks, NGN, are becoming increasingly open, shared and with
infrastructure that is reliant on highly distributed components. This is largely be-
ing driven by the vision of ubiquitous broadband access that is continually evolving
the way business and consumer customers interact. These networks must thus con-
tinue to improve in terms of performance through multiple dimensions, including
for example service mobility, personalization, transparency and immediacy.
This evolution of network infrastructure offers operators the possibility to
create many new forms of business. However, it also poses some significant new
challenges in many areas of communications and service management, especially
in resource-limited access networks. The NGN view is to rely upon an all-IP in-
frastructure, offering a clean separation between network and service layers and
enabling QoS provisioning “out of the box”, which should be easier to manage and
less expensive to maintain. However there are several factors which complicate the
overall NGN picture.
To be published in Calisti, M., Meer, S. and Strassner, J. (eds.) Advanced Au-
tonomic Networking and Communication - Whitestein Series in Software Agent
Technologies and Autonomic Computing.
2 M.Calisti,R.GhizzioliandD.Greenwood
End users are increasingly demanding new services, and dynamic, case-specific
service aggregations, to support a seamless and consistent experience across mul-
tiple access technologies, devices and locations. They expect to be always best-
connected, i.e., to have anywhere and anytime access to the best available tech-
nology with the maximum capacity on offer, plus easy-to-use and problem-free
services, all at ever lower prices.
Indeed, the proliferation of applications, services and heterogeneous tech-
nologies, including advanced multi-modal end users devices, enables a variety of
ubiquitous deployment scenarios, but also poses significant challenges in terms
of service usability and personalization. This is further complicated by the need
to integrate new solutions with legacy systems, while optimizing resource-limited
consumption (e.g., radio frequency in access networks).
In addition, the widespread expansion in the availability of high-speed broad-
band access technologies including cable, DSL, powerline, satellite, and wireless,
is encouraging the entry of new service providers in both the fixed and mobile
telecom sectors, thereby stimulating a competetive environment. In response op-
erators need to identify means of lowering operating costs by optimizing service
provisioning performance and connectivity management.
It is our belief that a fresh approach is required to achieve these objectives. We
thus propose a comprehensive policy-driven, autonomic software solution spanning
provider infrastructure and end-user devices that positions auto-adaptive control
software directly within the devices. The ma jority of service and connection pro-
visioning appraches in use today tend to operate on the traditional client/server
model and are thus rather ineffective due to a common inability to handle the
increasing dynamicity and diversity of heterogeneous access technologies. In this
perspective, emerging solutions need to be “autonomic” by design; their compo-
nents should be able to self-regulate and dynamically optimize their own behaviour
according to detected changes in their host environment [1].
We call our approach the Living Systems Autonomic Service Access Manage-
ment suite, LS/ASAM. It is a comprehensive and innovative solution that enables
effective delivery of next-generation ubiquitous services by dynamically combining
end user requirements and service provisioning policies with network-facing man-
agement and control functionality. By automating selected low-level processes on
both the user and operator sides and introducing more “personal intelligence” (user
context and behavior awareness) and “network intelligence” (network services,
content and resources awareness) throughout the whole service delivery chain, the
LS/ASAM solution realizes Autonomic Service Access Management (ASAM). The
guiding ASAM vision is to use autonomic techniques that enable operators to
efficiently manage and optimize resource utilization, performance and end user ex-
perience. This is achieved by transparently tuning service parameters while taking
into account changes in both the client and network context.
This chapter continues with a discussion of the ASAM core principles before
presenting the architecture and features of the LS/ASAM Suite as a means to
Autonomic Service Access Management 3
realise the ASAM vision. We then describe two key deployment scenarios coupled
with a discussion of some of the most distinctive characteristics.
Subsequently we provide some data on experimental work conducted in the
laboratory on performance analysis of the LS/ASAM suite prototype. The ASAM
simulator is described before the presentation of selected results from recent ex-
periments.
We conclude the paper with some discussion remarks, experimental conclu-
sions and targets for ongoing work.
2. Autonomic Service Access Management
Due to the increasing deployment of multiple access technologies at the edges of
networks, the management of ubiquitous communications and services is changing
rapidly. Intelligence and specific management and control functions need to be
migrated toward the edge of the network and even onto the customers’ devices.
In particular, service access management, i.e., the set of functions including the
selection and maintenance of one of several available communication channels, is
increasingly demanding:
•Fast and appropriate adjustment of the relevant connectivity parameters to
a continuously changing network environment.
•The assurance of sufficient service quality and reliability, whose perception
can vary from one user to another.
•In coordination with the aforementioned points, the optimisation of resource
usage and reduction of operational costs.
Autonomic Service Access Management, ASAM, addresses these issues by
dynamically and automatically adapting the configuration and utilization of avail-
able network access resources in a reliable and cost-efficient way. This is achieved
by embedding specialized intelligence into complex multi-technology and multi-
service access networks, including end user devices. The chosen approach is to
deploy smart techniques allowing operators to efficiently manage and optimize re-
source utilization, performance and end user experience. This by transparently
tuning service parameters (e.g., bandwidth, average delay), while taking into ac-
count changes in the context, including user preferences, Service Level Agreements
(SLAs), user location, devices features, and network resources.
ASAM bases its adaptivity on the capability to autonomously observe, ex-
tract, understand and use context information to consequently modify its own
functionality. Information exchange and correlation between client devices and ac-
cess nodes, as well as between access nodes even of different technologies, is at
the core of this approach. In particular, through dynamic mediation between (of-
ten conflicting) requirements on the client and network side, capacity for given
connection requests is allocated by taking into account the status of the whole
service provisioning chain. This requires accounting for a variety of parameters
4 M.Calisti,R.GhizzioliandD.Greenwood
that characterize the connection to be created, the consequently required network
resources, and the policies existing both on the user and provider side.
For this to be realized, flexible and distributed monitoring, configuration and
maintenance tools need to be smoothly interfaced and integrated within the evolv-
ing networking environment and pre-existing management systems. This is not an
easy task, especially when considering that many operators must deal with a di-
verse mix of systems and processes that make it difficult to effectively monitor and
tune service performance once already in the delivery phase. In this perspective,
a new kind of management solution is needed. A comprehensive policy-driven and
autonomic architecture, spanning basic infrastructures and end-user devices, which
builds adaptive control functionality directly into the corresponding elements, en-
abling the shift of focus from technology to value-added services.
LS/ASAM is a comprehensive ASAM solution that addresses these challenges
by making use of software agent technology [2]. Autonomous agents that adapt to
changes in the environment, minimizing human intervention and service interrup-
tion, lie at the foundation of LS/ASAM and provide a powerful means to engineer
a distributed and autonomic system that includes:
•Customizable and adaptive routines for automating and tuning repetitive
information and control tasks.
•Coordination mechanisms enabling the spontaneous collaboration and dy-
namic aggregation of services.
•Abstraction of communication components to support context changes thro-
ugh adaptation of semantic grounding.
In this way, autonomous software agents acting as autonomic managers, see
Figure 1, are enabling LS/ASAM to exhibit self-management capabilities that
increase reliability and performance while reducing operational and management
costs. This shifts the burden of many support and control tasks from users to
the underlying solution, which assists, facilitates and empowers human decision
making.
More specifically, LS/ASAM is a middleware solution empowered with auto-
nomic self-management capabilities, including:
•Self-configuration: policy-based self-configuration of the Suites components
according to changes in their usage and working environment.
•Self-optimization: proactive monitoring and control of resource usage, per-
formance and end user experience to enforce optimal behavior.
•Self-healing: automatic fault discovery and correction, both on the end user
devices and network elements.
•Self-protection: automatic detection of and protection from unauthorized sys-
tem control changes.
Control over LS/ASAM components is expressed through policies bound to
user preferences and business goals. The system senses, analyzes, plans and exe-
cutes changes in the environment to ensure that business goals can be effectively
met.
Autonomic Service Access Management 5
Context-Awareness
Autonomic Component
Autonomic Manager
Analyze Plan
Monitor
Execute
Sensors Actuators
Managed Resources
Knowledge
Policies and Rules
Agent Boundaries
Figure 1. An autonomic component architecture.
Although other approaches have been proposed in the literature that ad-
dress part of the ASAM challenges, none, to our knowledge, is able to dynamically
mediate between network and client requirements and accommodate resource al-
location and consumption accordingly. In particular, the solution presented in [3],
which is the closest one to LS/ASAM, supporting vertical handover in radio access
networks. In this system, a dedicated decision module, placed within a concrete
provider system, can communicate with various network devices, including client
devices, to determine radio access network selection based on QoS parameters.
Some degree of negotiation takes place, but only between entities within the net-
work and excluding the client devices that remain passive.
3. The LS/ASAM Suite Architecture
The LS/ASAM architecture includes two main types of autonomic software com-
ponents, as depicted in Figure 2, which communicate by relying upon the use of
common interaction protocols and a shared semantics-based ontology defining all
LS/ASAM concepts. These components are:
•LS/CA, the Living Systems Connection Agent, is a client component that
can run on a variety of mobile end user devices (e.g., laptops, PDAs, smart
6 M.Calisti,R.GhizzioliandD.Greenwood
Access Node
Controller
End User
Device
Access
Node
Semantic-based
Communication
User Data, Control Data,
Standard Messages
LS/SAM
Java Runtime
Environment
LS/ASAM Ontology
LS/ASAM Comm.
Protocols
LS/SAM Logic
Knowledge Base
LS/CA
LS/ASAM Ontology
LS/ASAM Comm.
Protocols
Java Runtime
Environment
LS/CA Logic
Knowledge Base
Standard
LS/ASAM Messages
Figure 2. An overview of the LS/ASAM architecture.
phones) and provides mobile users with improved quality and reliability by
optimizing service access through adaptive connection handover across multi-
ple access technologies and dynamic mediation of service delivery parameters
on behalf of the end user.
•LS/SAM, the Living Systems Service Access Manager, is a network compo-
nent that can run on hardware located at the access nodes or at a network
management facility. It dynamically optimizes resource allocation across het-
erogeneous network access domains with adaptive problem recovery and load
balancing techniques.
These lightweight software components, i.e., they can live as processes in a
Virtual Machine, can flexibly complement and extend many existing service man-
agement architectures, and are able to run on resource-limited devices and support
asynchronous communication with intermittent network connections. By dynami-
cally coordinating their actions and behavior, they enable adaptive communication
service access by mediating between operator policies and end-users requirements
and preferences.
3.1. The Living Systems Connection Agent
The LS/CA component provides adaptive service access by setting connectivity
parameters according to the outcome of a mediation process to establish a service
Autonomic Service Access Management 7
access agreement based on the end user’s requirements and the network provider’s
offering. This is determined by a set of factors including:
•Quality requirements of the applications and services running on the device
the LS/CA is embedded in.
•Physical end user device status, e.g., battery power level, and properties, e.g.,
available network interfaces.
•Existing service provisioning conditions according to pre-defined subscription
contracts/SLAs.
The LS/CA proactively manages and processes this information according
to policies which capture end user preferences, e.g., minimising connection costs,
maximising battery life when on-the-move, etc., and supports the following main
features:
•Seamless handover and session continuity. This guarantees interruption-free
service access across multiple technologies by allowing an LS/CA empowered
device to maintain the same IP address for an entire session. This is achieved
by making use of Mobile IP technology [4].
•Secure communication. Tight integration of the LS/CA with several third
party VPN clients allows permanent secure connectivity. Furthermore, by in-
tegrating IPSec [5] and Mobile IP, the LS/CA ensures end-to-end encryption
of all generated traffic (as an optional feature).
•Connection adaptation. This indicates automatic detection of available net-
works and selection of the preferred network adapter (access technology)
based on service requirements and network conditions for improved reliabil-
ity and QoS. This can trigger dynamic mediation between the LS/CA and
the LS/SAM components.
•Context-aware user support. Through semantic service specifications, policy-
driven decision making and dynamic information retrieval, the LS/CA im-
proves end-user experience by directly addressing low-level issues (e.g., failure
recovery, connection adaptation), while taking into account user policies and
boundary constraints, i.e., context-based information and coordination with
LS/SAM components as needed.
From the LS/CA perspective the mediation process is initiated by sending a
Call For Proposal, CFP, to one or several LS/SAMs. Naturally any LS/SAM with
a open connection established with the LS/CA may also receive the CFP so that
it can also participate in the connectivity mediation process.
3.2. The Living Systems Service Access Manager
The LS/SAM component proactively monitors traffic and resources in the access
node it controls, triggers appropriate actions (e.g., vertical handover, load bal-
ancing) according to the network status and current traffic conditions, processes
incoming LS/CA calls for proposal and elaborate offers as appropriate - see Sec-
tion 3.3. In particular, the two main distinctive features enabling LS/SAMs to
optimize resource consumption at the access network level are:
8 M.Calisti,R.GhizzioliandD.Greenwood
•Load-balancing. Balancing traffic load across WLAN and cellular networks
while considering the QoS needs of running services renders the network
more resilient to traffic peaks. This is achieved by dynamic coordination
between LS/SAMs that can hand over a certain number of connections to
neighboring access nodes according to possibly several operator policies. The
use of distributed constraint satisfaction algorithms [6] for LS/SAMs peer-
to-peer orchestration enables effective load balancing by taking into account
all existing constraints.
•Congestion recovery. Real-time and proactive detection, analysis and relief of
congestion, reduces call dropping and increases service resilience and avail-
ability. Within an access node, once no new network connection can be ac-
cepted or the total requested bandwidth exceeds the total available one, i.e.,
packets are dropped, an LS/SAM can decide upon specific policies and ex-
isting SLAs (if any) whether and how to drop or hand over part of the traffic
to neighboring access nodes.
LS/SAMs decisions and behavior are guided by the operator’s policies that
express service provisioning preferences with respect to a variety of aspects includ-
ing, e.g., how to allocate traffic to balance out network utilization, how to treat
specific users (i.e., connections) in case of congestion, how to adapt pricing schemes
according to the user’s subscription type. This requires dynamic management of
information including:
•Traffic conditions and resources available within the access node the LS/SAM
is controlling.
•Traffic conditions and resources available in other access nodes that a given
portion of traffic can be handed over to, via dynamic LS/SAM-to-LS/SAM
coordination.
•Existing service provisioning conditions according to pre-defined subscription
contracts/SLAs.
3.3. Adaptive Coordination of the LS/ASAM Components
The mediation process conducted between the LS/CA and LS/SAM components
consists of a sequential interchange formulated as a contract-net protocol [7] ne-
gotiation with the goal of determining the best connection parameters given the
requirements of the end user, the offering of the network provider and the condi-
tions of the transmission medium.
The requirements of the end user toward the provider are a combination of
(i) the preferences of the end user formulated as user policies (e.g., minimising
connection cost), (ii) the quality demands of the applications running on the end
user device (e.g., a given application may require low end-to-end delay), (iii) the
status of end user device resources (e.g., battery power, which can affect the se-
lection of the transmission technology), (iv) the technologies supported by the
end user device (e.g., only WLAN and UMTS network interfaces available), and
Autonomic Service Access Management 9
loop
LS/CA LS/SAM
call for proposals
accept proposal
initiate service set up
[proposal = accepted OR iteration = maxallowed]
make proposal
reject proposal/
counter-propose
formulate proposal
(or ignore)
Loop for re-
negotiating
proposals
consider
proposal
Figure 3. Mediation process between the client and network
LS/ASAM components.
(v) the conditions stated in the subscription contract (e.g., costs for using certain
technologies).
The offering of the provider toward the end user is determined by considering
(i) the properties of the provider network (e.g., diversity of network access tech-
nologies), (ii) the network status (e.g., distribution of traffic load, delay times), (iii)
the capabilities of the network (e.g., mobility support, QoS control) and (iv) the
provider policies, including business rules, that relate to the use of its infrastruc-
ture, pricing schemes, traffic prioritization mechanisms, etc.
Figure 3 illustrates the typical message exchange during a proposal setup
sequence. An LS/CA sends a Call For Proposal (CFP) to one or several LS/SAMs
requesting offers to set up a connection with specified constraints including quality
requirements, or connection characteristics.
An example of a simple CFP is:
(set up connection, (min. bandwidth: 100 KBit/s.
max. delay jitter: 50 ms))
Once sent to all prospective LS/SAMs, the LS/CA waits until some predefined
deadline to receive proposals and/or rejections. Any LS/SAMs that have not sent
a proposal or rejection by this deadline are considered to have been unable or
10 M. Calisti, R. Ghizzioli and D. Greenwood
unwilling to respond to the CFP. A simple example of a proposal sent by a re-
sponding LS/SAM is:
(set up connection, (network: UMTS, min. bandwidth: 100 KBit/s, max.
bandwidth: 120 KBit/s, max. delay jitter: 40 ms, max. end-to-end
delay: 200 ms))
This proposal includes some additional connection parameters than those present
in the orginal CFP. Although not mandatory to do so, these can be taken into
account by the the LS/CA when evaluating the suitability of the proposal.
The proposals are assessed by means of the Proposal Assessment Function
(PAF) that takes as input (i) the set of quality requirements stated in the original
CFP, (ii) the received proposal (or the relevant parameters stated in the proposal),
(iii) optionally, the user preferences (that can be formulated as user policies), (iv)
optionally, the status of the end user device (e.g., battery power level that can
affect the selection of the transmission technology), (v) optionally, the properties
of the end user device, (vi) optionally, the capabilities of the end user device and
(vii) optionally, any Quality of Experience, QoE, metrics, (vii) optionally, the set
of network operator policies including business rules.
The PAF computes a sum of weighted differences between the required quan-
titative parameters and their corresponding values in the proposal. Nominally, the
PAF is normalised to a target value domain 0,1 where 0 indicates that the proposal
does not satisfy any requirements and 1 indicates that the proposal is valid and
fully acceptable. Intermediate results between these bounds indicate the degree to
which the proposal meets the CFP requirements. Ancilliary annotations record if
the proposal exceeded the CFP requirements for use with counter-proposal nego-
tiations.
At this point the LS/CA must decide whether to make a counter-proposal
to any number of selected LS/SAM’s that responded favourably to the original
CFP. This decision is made in accordance with how well a received proposal meets
or exceeds the original CFP request. If selected, a counter-proposal can be is-
sued to a responding LS/SAM in an attempt to initiate bilateral negotiation to
revise the proposed offer. Multiple counter-proposal negotiations can be handled
concurrently by an LS/CA with active PAF based comparison of each to deter-
mine variances between returned proposal updates thereby assisting with refining
individual negotiations by taking into account all ongoing negotiations.
A counter proposal is created by modifying a received proposal in accordance
with preferred characteristics. If the original CFP sent was:
(set up connection, (min. bandwidth: 100 KBit/s))
With a received proposal being:
(set up connection, (min. bandwidth: 70 KBit/s))
Autonomic Service Access Management 11
The PAF determines that this received proposal is close to its requirements, as
expressed in the original CFP, and thus creates a counter proposal in order to initi-
ate fine-grained bilateral negotiation with the sender of the proposal. The counter
proposal in the instance of this example may be that the proposed 70Kbit/s band-
width offer is iteratively increased to 80KBit/s:
Counter proposal: (set up connection, (min. bandwidth: 80 KBit/s))
This counter proposal is a compromise between the original bandwidth spec-
ified in the CFP and the bandwidth offered in the returned proposal.
It is important that the decision process exhibits a convergent behaviour to
avoid continuous proposal revision. Several suitable algorithms can be found in
the literature include that by Hofbauer et al. [8] and by Shamma et al. [9].
When, or if, a proposal is accepted the client device sends an accept-proposal
message to the corresponding network provider. All other proposals that have been
received are explicitly rejected by informing their source providers. The reason for
rejection may be included in the message.
3.4. Technology Foundation
As networks grow increasingly larger and more complex, they become harder to
manage efficiently and reliably. This is even more challenging in resource-limited
access networks, which affects the capability to deliver true seamless mobility.
Thus, network and service management solutions are required to exhibit autonomic
behavior.
Their components detect, diagnose and repair faults, adapt their configu-
ration and optimize their performance, while protecting and healing themselves
according to changes in the network and operating environment.
The key idea is to assist, facilitate and empower humans (operators, network
administrators, customers) by shifting the burden of many support and control
tasks from them to the underlying solution components.
As anticipated in Section 2, the LS/ASAM Suite has been conceived and
realized by embedding autonomic self-management capabilities at the core of its
functionality. Its components autonomously observe, extract, understand and use
context information to consequently modify their functionality, according to poli-
cies that are bound to business goals. The autonomic capabilities of the client
components, LS/CA, and the network component, LS/SAM, are classified as fol-
lows:
Self-configuration. The LS/CA adjusts its own configuration according to changes
in the working environment in which the user device is located. Policy-controlled
profiles for different locations identify the configuration of features to be used,
e.g., connection type, VPN, file shares. The LS/SAM performs self-configuration
determining its own behavior to achieve high-level directives. This enables the
network (namely the access resources the LS/SAMs control, e.g., base stations
or access points) to respond dynamically to changes in operator policies and/or
12 M. Calisti, R. Ghizzioli and D. Greenwood
network state. Different load balancing strategies may be adopted, depending on
traffic conditions, resource availability and SLAs.
Self-optimization. The LS/CA selects a specific connection type according to user
policies and in relation to changes in the context. This is particularly beneficial
while roaming in partner networks where the nominal connection may not be
the preferred, best or indeed cheapest option. The choice of alternative network
adapters can also be triggered by the need of optimizing specific application perfor-
mance in relation to device properties and network status. The LS/SAM efficiently
manages access node resources to meet specified performance objectives under dy-
namic operating conditions. By proactively balancing load across distinct access
nodes (via interaction with peer LS/SAMs) and triggering vertical handover of se-
lected connections, it is possible to optimize network performance and availability
according to existing operators policies.
Self-healing. The LS/CA detects faults in related system components (e.g., net-
work cards, drivers, system interrupts) and transparently takes action to repair
and circumvent the anomalous behavior. The LS/CA also attempts to re-establish
lost connections or, if not possible, seamlessly transitions to a session over an al-
ternative connection type. The LS/SAM is able to detect and repair unpredictable
conflicts between service requirements and available network resources. If appro-
priate, it coordinates its behavior with other LS/SAMs. In particular, real-time
and proactive detection, analysis and relief of congestion allows the LS/SAM to
reduce call dropping and thereby increase service resilience and availability.
Self-protection. The LS/CA detects unauthorized alterations to obfuscated oper-
ator policies stored in the system registry. It stalls operations while replacing the
policies with securely obtained replacements. The LS/SAM performs the necessary
traffic analyses to detect potential security threats and informs peer LS/SAMs, the
overall network management system and/or the network administrator. In partic-
ular, the LS/SAM supports identification of malicious nodes that attempt denial
of service attacks and blacklists them, warning the complementary access network
management components.
4. The LS/ASAM Suite in Action
Ubiquitous data connectivity and communications management are optimised
transparently across multiple network access technologies by dynamic coordination
of the LS/ASAM components according to the specific situations. In particular,
different combinations of their features enable a variety of deployment scenarios. In
the following, two of the most significant ones are presented including a discussion
of the distinctive characteristics in relation to relevant work.
4.1. QoS Enforcement in Heterogeneous Access Networks
The notion of guaranteed data transmission quality with enforcement mechanisms,
in particular for emerging QoS sensitive multimedia applications, e.g., voice or
video over IP, is a key issue especially in converged networks [10]. While traffic
Autonomic Service Access Management 13
prioritization is often not of paramount importance in core networks due to over-
provisioning, QoS is an essential differentiator in limited-capacity wireless access
networks for capacity and/or delay sensitive traffic such as voice or video over
IP. While for cellular access technologies belonging to 2.5G, 3G and 3.5G, appro-
priate standards for QoS have been defined, few operators yet make widespread
use of them. In addition, the WLAN world is supporting its technologies with
specifications that directly account for QoS management.
In particular, when integrating different access network technologies, e.g.,
WLAN and UMTS, the quality of a connection may be degraded during vertical
handover where (i) the connection needs to be re-established at the new access
node, which is time consuming and during which no data can be transmitted, and
(ii) if too many IP packets are lost, they must be retransmitted which can also be
time consuming in the case of a large number of packets - again leading to service
interruption.
Various approaches have been developed and proposed to address this prob-
lem. In [11], a reservation-based QoS model for integrated cellular and WLAN net-
works is defined and an adaptive mechanism to ensure end-to-end QoS is proposed.
However, this model can only work by making the assumption that cellular/WLAN
interworking is realized by relying upon a common and uniform reservation-based
QoS architecture, which is not (yet) the case for most real network scenarios. Sim-
ilarly, Song et al. [12] proposed an admission control mechanism for integrated
voice and data services in cellular/WLAN networks. The main limitation of this
approach though is that it does not account for video traffic.
To effectively provision QoS and optimize resource utilization for a variety
of possible heterogeneous network scenarios, the LS/ASAM Suite relies upon the
dynamic combination of specific mechanisms both at the client side (i.e., seamless
handover, session continuity and connection adaptation) and at the network side
(i.e., congestion recovery and load-balancing) that are compliant with dominant
industrial standards, e.g., mobile IP or SIP/IMS, when supported, or technology-
independent, whenever possible.
Unlike legacy systems and hardware-based solutions, the LS/ASAM compo-
nents accommodate high-level service and user needs and preferences (including
QoS requirements) by implementing coordination mechanisms and resource alloca-
tion algorithms that hide low-level access technology dependent processes. This is
achieved by deploying an agent-based middleware architecture that provides users
with a common and higher level of abstraction, which makes low-level network
access heterogeneity transparent.
On the client side, basic QoS in terms of service availability and continuity
is enforced by the LS/CA through automatic and policy-driven vertical handover,
i.e., all traffic is switched from one network interface, according to existing con-
straints and user policies. Moreover, by continuously monitoring network condi-
tions and device status and properties, the LS/CA exerts QoS and context-aware
resource management by selecting the most appropriate access technology to be
14 M. Calisti, R. Ghizzioli and D. Greenwood
used for the running applications/processes. In addition, when appropriate, as de-
tailed in Section 3.3, the LS/CA can also trigger negotiation with one or more
LS/SAMs for different connectivity conditions.
On the network side, the key mechanisms deployed by the LS/SAM to enforce
QoS provisioning are load-balancing and congestion recovery. Load-balancing can
be triggered by LS/SAMs in order to redistribute traffic across several access nodes
according to various criteria, including:
•Current utilization of resources at the access node, e.g., once the traffic over-
comes a given threshold a certain portion of the supported connections might
be handed over to neighbor LS/SAMs.
•QoS requirements of the running services, e.g., best-effort connections might
be handed over to prioritize premium services for which charging might be
based on service reliability guarantees (e.g., ≥95% non-disruption).
•Predictions of the network resources usage to minimize the probability of
congesting an access node.
Analogously, whenever congestion occurs a specific part of the traffic at a given
access node might be handed over to other LS/SAMs or selected existing connec-
tions (e.g., the non-premium ones) might even be dropped as appropriate. This
enables relief of congestion and increases service resiliency and availability.
For example, assume a user that launches an IP-based TV program (e.g., a
news channel) on a smart phone. During the launch of the selected application
to render the video stream, the LS/CA determines the connectivity parameters
(typically bandwidth and delay) for interruption-free high quality service provision.
Because different access technologies offer different QoS assurances, the LS/CA
might try to switch to a specific technology, e.g., UMTS, that better supports
the QoS level needed for the video down-streaming. In addition, in the case of an
UMTS connection, the LS/CA would set up a new Packet Data Protocol context
requesting the UMTS QoS streaming class [13].
Figure 4 depicts the deployment model for this case. Each end user device
is installed with an LS/CA component able to enforce QoS. The LS/CA must be
aware of the different traffic categories available in each network access technology.
During a vertical handover, the QoS class of the active network is mapped into an
appropriate QoS class of the target network. There is one LS/SAM agent being
deployed per access node, i.e., each LS/SAM agent is in charge of a specific access
node and thus is up-to-date at all times regarding the status of that node. When
planning load balancing and congestion recovery, the LS/SAM agent must be aware
of the QoS classes supported by the different access technologies to minimize the
risk of degraded service quality. This involves LS/SAM-to-LS/SAM coordination
first to exchange information on current traffic load (or resource availability) and
then to possibly take or hand over part of the communications/traffic1.
1Peer LS/SAMs coordination is not described in this paper because of some pending patenting
issues.
Autonomic Service Access Management 15
QoS enabled
UMTS base station
QoS enabled
WLAN access point
Core Network
LS/CA
LS/SAM
LS/SAMWLAN
LS/SAMUMTS
LS/CA
QoS information
QoS information
exchange in
preparation of
handovers
Agent Control
Signalling
LS/ASM Layer
Applications
Web Video
over
IP
QoS requirements
(video traffic)
No specific QoS
requirements
(best-effort traffic)
Logical Integration
Points
Radio Link
Component
Software Agent
End User
Device
Legend
Figure 4. Deployment model of the LS/ASAM Suite for QoS enforcement.
4.2. Integration with an IMS/SIP Framework
IP Multimedia Subsystem, IMS, initially developed by 3GPP and 3GPP2 as an IP
core network architecture for cellular/wireless-based access to Internet services, is
now evolving into a standard that provides a common framework to create and
offer next generation converged network services [14]. IMS builds on the Session
Initiation Protocol, SIP, that is mainly in responsible for delivering a session de-
scription to a user at its current location [15]. The key idea is to enable any kind
of access (wireless or fixed) for any kind of media (including any combination of
voice, text, image and/or video) supporting multiple devices and endpoints.
Because of the (at least initial) co-existence of IMS and non-IMS applications,
the costs associated with moving to a full IMS-based network, and the inherent
complexity of IMS (and its several standards, interfaces and protocols) most ser-
vice providers and or operators are expected to migrate toward an IMS service
framework iteratively.
One of the core issues to be addressed for successful adoption of IMS is the
ability to face more aggressive bandwidth and latency demands, which implies
increased QoS management and design capabilities on the bearer network [16]. In
particular, IMS/SIP lacks traffic management capabilities and especially adaptive
connectivity management and optimization mechanisms that can be regarded as
key components for delivering ubiquitous quality-sensitive multimedia services.
In this perspective, the LS/ASAM Suite complements an IMS-based frame-
work by ensuring the quality of delivered services at the bearer network level
16 M. Calisti, R. Ghizzioli and D. Greenwood
through its adaptivity mechanisms, leaving IMS/SIP to cope with call control and
service deployment issues. As depicted in Figure 5 the LS/CA component directly
interacts with the SIP client installed on the end user device. In this way, the SIP
client is able to obtain information on the quality of the connection which is helpful
to determine, for instance, the appropriate codec to use, and to request the LS/CA
component to ensure a certain quality level (in particular, when explicit QoS class
enforcement is enabled). On the network side, an LS/SAM agent integrates with
each access node and, by means of load balancing and congestion recovery enables
to provide a high level of service quality.
A simple use case is when one considers the collaboration between a SIP
client and the LS/CA component to guarantee a level of quality required by a user
to perform a video call (or, similarly, to watch Mobile TV). Upon launch of the
SIP-based video calling application, the SIP client assesses the connection quality
by means of the LS/CA component. The SIP client is aware of the quality require-
ments imposed by the video call service that are also variable according to the
size and quality of the video picture. The LS/CA component can, in collaboration
with the respective LS/SAMs, discover the quality offering at alternative access
nodes and, based on that decide whether a handover to another access node needs
to be triggered. Both end devices that participate in the video call must also agree
on the codecs to be used for encoding and decoding the voice and video data.
The LS/CA component delivers the necessary information to the SIP client to
make its choice. Once the video call is established and running, it is the LS/CA
agent’s responsibility, in cooperation with the active LS/SAM agent, to preserve
the quality of the connection and take appropriate measures if tolerance thresh-
olds are violated. Depending on the mobility profile of the user, but also on the
evolution of the network conditions, handoffs are unavoidable and thus need to be
well planned and efficiently executed to minimize quality breaches.
The LS/CA does not affect the SIP call itself nor infringe any of the IMS/SIP
standards. SIP is concerned with controlling the call execution while LS/ASAM
takes care of connectivity. LS/ASAM is therefore complementary to IMS/SIP and
benefits result even if only a small proportion of the entire network infrastructure
(namely the access part) and end user devices are LS/ASAM empowered.
5. Experimental Analysis
In order to give a measure of the concrete benefits brought to a telecom opera-
tor by the adoption and deployment of a solution based on LS/ASAM, several
experimental tests have been performed. This section first introduces the ASAM
simulator, an instrument built for validating the basic concepts and evaluating
various autonomic service access strategies on a set of simulated network settings
representing real scenarios. One particular scenario is then selected to illustrate
performance when different service access algorithms have been deployed in the
Autonomic Service Access Management 17
IP Core Network
End User
Device
RNC
LS/SAM
LS/SAM
LS/SAM
LS/SAM
UTRAN
WLAN
SGSN
Quality
negotiations
GGSN
IP Router
CSCF
P-CSCF
I-CSCFS-CSCF
LS/ASM
Layer IMS
Layer
LS/CA
LS/CA
Terminal
Applications
SIP
Proxy
UE
Quality negotiations
Other control messages
SIP Applications
SIP Client 3rd Party
Application Servers OSA
Service
Platform
External IP
Network
CSCF: Call State Control Function
GGSN: Gateway GPRS Support Node
OSA: Open Service Access
RNC: Radio Network Controller
SGSN: Serving GPRS Support Node
UE: User Equipment
UTRAN: Universal Terrestrial Radio Access Network
SIP Signalling
Data Transfer
Legend
Logical Integration Points
Component
Software Agent
Figure 5. Deployment model of the LS/ASAM Suite when in-
tegrating with an IMS/SIP-based architecture.
user devices and in the access network. A set of preliminary experimental results
are provided, obtained from the comparison of the discussed access strategies.
5.1. The ASAM Simulator
The ASAM simulator is an instrument built in Java for validating the ASAM
concepts, in particular, how different autonomic access strategies deployed into
LS/CA and LS/SAM modules should perform in real network access scenarios.
In the real world people use their portable devices to request services in ac-
cordance with changes in location, activity, and other requirements. Using radio
communication they are able to connect to a network operator offering a hetero-
geneous infrastructure of different access node types (e.g., WLAN access nodes,
GPRS/UMTS antennas, etc.) In the ASAM simulator, both user devices and ac-
cess network components are modeled using software agents which. Agents that
simulate a user device can make use of the LS/CA where specific service access
strategies are pre-loaded. In the same way, an agent representing an access network
component can make use of the LS/SAM capabilities. The interaction between a
device and an access node is then mapped through an exchange of FIPA-compliant
messages.
18 M. Calisti, R. Ghizzioli and D. Greenwood
Input Parameters. Within the ASAM Simulator time is discrete and the simula-
tionsarebasedonthequasi-static condition. For this reason, input parameters
related with the time are:
•Start time of the experiment.
•Duration of the experiment.
•Duration of a time step (e.g., 1 minute).
Furthermore, other parameters are required to describe the scenario:
•Locations represented in the experiment (e.g., train station, street, offices,
etc.).
•Types of available network interfaces (e.g., UMTS, EDGE, etc.).
•Set of network services that are simulated in this experiment (e.g., phone
call, VOD, email, etc.).
•Set of access nodes.
•Set of end user devices.
For each access node (e.g., WLAN access point, UMTS cell, etc.) the following
input parameters are required:
•Type of network technology represented by this access node (e.g., UMTS
cell).
•Nominal bandwidth of an access node measured is Bit/s.
•Maximum number of concurrent connections.
•Maximum bandwidth deployable on a single connection.
•Version of the LS/SAM the access node makes use (e.g., none, LS/SAM-
BN 2).
Finally, for each user device to be simulated, the following input parameters are
necessary:
•Location of the device at the beginning of the simulation (e.g., street).
•Set of network adapters installed in device (e.g., only GPRS).
•Amount of bandwidth that can be used at maximum given the network tech-
nology (e.g., 11Mbit/s for WLAN).
•Set of service descriptors denoting the services that are available to the user
who operates the device (e.g., a normal mobile phone can perform only calls).
•Version of the LS/CA the user device makes use (e.g., none, LS/CA-APF2
).
•A set of input parameters used to define a mathematical model which de-
scribes the behaviour of the end-user while using the device. This is defined
in terms of movements among locations, usage rate and duration of services
while being located in a given place. In particular, the following matrices
must be provided:
–The average time before an user changes her location, moving from one
environment to another one.
2The suffix (APF in this case) determines the type of autonomic access strategies the component
implements.
Autonomic Service Access Management 19
–The average time before an user issues a service demand while being
located in a given space.
–The duration of a started service while being located in a given space.
The implemented mathematical model is based on the Markovian property
that the probability of the occurrence of an event does not depend on the
history of previous events. Based on this property, events like the starting of a
service or the movement between locations are simulated with an occurrence
rate equal to the inverse of the λparameter of a negative exponential distri-
bution. Furthermore, the duration of of a started service is simulated using
the Erlang-k distribution. The expected average and standard deviation of
the service duration are used to define the distribution.
In the ASAM simulator, each device has a user event generator that
implements this mathematical model. The generated user events represent
movements or service initiations with a stochastic duration. When an event
occurs, the action is simulated (e.g., start a VOIP call in a road for 2 minutes).
Each time an event is consumed, a new one is immediately generated. The
generator also terminates elapsed services.
Output Variables. The ASAM simulator provides a set of output parameters that
measure the performance of the LS/CA and LS/SAM strategies. The following list
of output parameters includes only the subset of those used in Section 5.3:
•Mur: The average used bandwidth of an access node in relation to its nominal
bandwidth. High Mur values encounter a high average utilization of the access
nodes which means that the infrastructure is more efficiently utilized.
•Msr: The satisfaction rate of a demand is an indicator for the service quality
that a user receives. Currently, this variable considers only the amount of
bandwith consumed versus the amount of requested.
•Mfc: The accumulated time span during which an end user device receives
the bandwidth it requests and thus can deliver full service quality to the user.
Values are normalized in the range [0..1].
•Md.vho: The average occurrence rate of vertical handoffs in a time step when
triggered by a user device.
•Mn.vho: The average occurrence rate of vertical handoffs in a time step when
triggered by an access node.
5.2. Simulation Setup
This section presents preliminary laboratory experiments conducted to validate the
ASAM concepts through the use of the ASAM simulator. The presented simulation
evaluates what might happen in a normal working day during which a large number
of people arrive at a train station before dispersing to their places of work where
they spend most of their day.
Figure 6 illustrates the simulated access network topology where different ac-
cess nodes (UMTS/GPRS cells and WLAN access points) cover different locations
20 M. Calisti, R. Ghizzioli and D. Greenwood
GPRS
GPRS
UMTS
UMTS
UMTS
WLAN
WLAN
Train Station Road 2 Office 2
Road 3
Road 1
Office 1
Figure 6. Access network topology used in the presented simulation.
Ta b l e 1 . Access nodes properties.
Type Nominal Bandwidth Max Bandwidth/Device Max. Connections
WLAN 2000 Kbit/s 2000 Kbit/s 120
UMTS 1500 Kbit/s 300 Kbit/s 6
GPRS 400 Kbit/s 50 Kbit/s 10
Ta b l e 2 . Average movement rate exhibit by the users.
PPPPPPP
P
From
To Train Station Road 1 Road 2 Road 3 Office 1 Office 2
Train Station -5mins 5mins - - -
Road 1 10 mins - - - 5mins -
Road 2 10 mins - - - - 5mins
Road 3 - - - - 10 mins 10 mins
Office 1 -5hours -4hours - -
Office 2 - - 5hours 4hours - -
(a train station, three roads and two offices). The access nodes exhibit the charac-
teristics presented in Table 1. The reported values are similar to the characteristics
offered by typical network components deployed in most access networks.
In this simulation, 40 users, starting from the train station, move around this
scenario with their devices consuming network services. Their devices are able to
handle communication with all the available network technologies (WLAN, UMTS
and GPRS).
Table 2 describes the average movement rates exhibited by the users. It is
important to notice that these rates are unidirectional, that is, the frequency of
Autonomic Service Access Management 21
Ta b l e 3 . Occurrence rate and duration of services started in
specific locations.
eMail
80Kbit/s
VOIP
128Kbit/s
Internet
240Kbit/s
VOD
1Mbit/s
Train Station 40 mins 2hours 3mins -
5mins±12mins±15mins±1
Road 1 1hour 2hours
5mins±12mins±1
Road 2 1hour 2hours
5mins±12mins±1
Road 3 1hour 2hours
5mins±12mins±1
Office 1 20 mins 1hour 1hour 4hours
5mins±110 mins ±115 mins ±230 mins ±5
Office 2 20 mins 1hour 1hour 4hours
5mins±110 mins ±115 mins ±230 mins ±5
moving from one location to another is not necessarily the same of the reverse
direction between the same locations.
The information about how services are used is described in Table 3. In
particular, the table presents the average occurrence rate and the duration of a
service, given the user location. Moreover, the service duration is described in
terms of the average time and its standard deviation. Additionally, the bandwidth
consumed by each type of service is defined in the table column headers. Values in
Table 2 and Table 3 do not represent a specific case but we believe these quantities
are sufficient to analyze how the selected access strategies behave.
The described scenario was simulated for 24 hours with a time step of 1
minute. Given the non-deterministic property of the experiments, 10 simulation
runs of the same scenario have been performed.
Network access strategies. In this experiment we consider two network access
strategies: one implemented by the LS/CA and one implemented by the LS/SAM.
The access strategy implemented in the LS/CA equipped user devices, named
Adapter Priority Fuction (APF), assigns a dynamic priority function at each
adapter of the user device. This function takes as input measured and expected
parameters values that are: battery power, time since last handover, used band-
width, end-to-end delay, adapter statistics, adapter cost and creates a weighted
linear combination of a set of sub-fuctions built on the listed parameters. If the
adapter with the highest function value is different then the current one, am han-
dover is triggered. We named an LS/CA that implements the APF access strategy
as LS/CA-APF.
22 M. Calisti, R. Ghizzioli and D. Greenwood
Ta b l e 4 . Comparison of the results obtained simulating four dif-
ferent network access configurations.
Without
LS/ASAM
Only LS/CA-
APF
Only
LS/SAM-BN
With
LS/ASAM
Mur 0.2821±0.0062 0.2897±0.0265 0.4048±0.0066 0.4199 ±0.005
Msr 0.5433±0.0070 0.5839±0.0637 0.7382±0.0137 0.7621±0.0138
Mfc 0.1153±0.0088 0.1540±0.0752 0.2216±0.0156 0.2583±0.0260
Mn.vho 0 0 0.1685±0.0212 0.1478±0.0064
Md.vho 0.0169±0.0008 0.0066±0.0006 0.0095±0.0010 0.0015±0.0001
The access strategy implemented in the LS/SAM equipped access nodes,
named Balance (BN), tries to keep high the quality of the services users require
balancing the load among the access nodes available in the user device’s neigh-
borhood. Whenever an established connection obtains less bandwidth than the
requested one, the access node using a Contract-Net protocol asks to other nodes
how much bandwidth they could offer to that connection. The candidate access
node should be able to satisfy the requested bandwidth and minimizes the gap
between the bandwidth demand and the bandwidth offered. If there are no access
nodes that offer more than the requested bandwidth, the connection is assigned
to that access node with the highest bandwidth offered. If no proposals are better
than what the current access node offers, no handover is performed. We name an
LS/SAM that implements the BN access strategy as LS/SAM-BN.
If both the LS/CA and the LS/SAM are deployed in the network, that is,
the whole LS/ASAM system is in use, a mechanism to avoid conflicts between
provider and user strategies is adopted.
In order to understand the benefits provided by LS/ASAM, the scenario
where LS/ASAM is not present was also simulated. In this case, access nodes do
not exhibit any access logic and the devices select the preferred access node based
on the highest nominal bandwidth a network technology provides (e.g., WLAN,
UMTS, GPRS).
5.3. Results
Table 4 presents the results obtained simulating four different network access con-
figurations: the case without the LS/ASAM system, the case with only LS/CA-
APF components, the case with only the LS/SAM-BNs, and the last case where
the whole LS/ASAM system is deployed.
The results show that if only LS/CA components are deployed in the network
(third column), they are able to improve all the evaluated metrics when compared
with the case where no LS/ASAM components are in place. For example, LS/CAs
generate 33% more time where users receive the requested bandwidth even with a
lower number of vertical handovers.
Results are even better if we compare the case without LS/ASAM to the case
where only LS/SAMs are deployed. In this case, for example, the user satisfactory
Autonomic Service Access Management 23
20 30 40 50
Number of User Devices
M
ur
0.00 0.10 0.20 0.30
Without LS/ASAM
Only LS/CA−APF
20 30 40 50
Number of User Devices
M
ur
0.0 0.1 0.2 0.3 0.4
Only LS/SAM−BN
With LS/ASAM
20 30 40 50
Number of User Devices
M
ur
0.0 0.1 0.2 0.3 0.4
Without LS/ASAM
Only LS/SAM−BN
20 30 40 50
Number of User Devices
M
ur
0.0 0.2 0.4
Without LS/ASAM
With LS/ASAM
Figure 7. The benefits provided by LS/ASAM (or some of its
components) against an increasing number of network users.
metric, Mfc, improves by 100% and the usage rate metric, Mur, improves by 42%.
From this we can conclude that an operator may be able to mitigate the need for
extensions to network infrastructure in lieu of deploying some software intelligence
into existing infrastructure.
Moreover, we can observe from these results that adding intelligence in the
access network brings about greater benefits than adding intelligence to user de-
vices. This is because the network has a broader and real knowledge of the current
infrastructure status than a user device that also bases its decisions on estimated
values.
Furthermore, we can notice that the combined use of LS/CA and LS/SAM
components (the whole LS/ASAM system) generates yet greater benefits than
those one generated by one of the two components in isolated use. We believe that
further improvements of the evaluated metrics will be obtained with as yet to be
reported work regarding the simulation of collaborative strategies between LS/CA
and LS/SAM as presented in Section 3.3.
Finally, Figure 7 shows the benefits provided by LS/ASAM (or some of its
components) with the increasing of the network users when analysing the Mur
metric, that is, usage of the network. Considering for example the histogram in the
bottom-right of the figure, where a system with LS/ASAM deployed is compared
to one without, it is notable that as the volume of users increases, so does the gain
achieved with LS/ASAM.
To measure the real significance of the obtained results the Wilcoxon Paired
Rank Sum Test was applied. This test stated with a confidence level higher that
24 M. Calisti, R. Ghizzioli and D. Greenwood
95% that the improvements generated by the LS/ASAM system are statistically
significant.
6. Discussion and Conclusions
The LS/ASAM Suite is a distributed and resilient system that exhibits high adap-
tivity to its network environment. This has been achieved by properly combining
multi-agent systems concepts and technology with powerful resource allocation
algorithms and reasoning strategies.
The central idea is that loosely-coupled distributed management functions
and control methods can be well-modeled and implemented by making use of
automated, goal-driven and proactive software entities. These lightweight com-
ponents are able to operate on resource-scarce devices and support asynchronous
communication with intermittent network connections. Moreover, according to the
results of proactive monitoring information received from the environment within
which they are embedded, the LS/ASAM components directly assist with auto-
nomic management of network resources. They are able to configure themselves
and dynamically optimize their operations according to the way their environment
changes and in-line with operator and client user policies. They thus assist with
the speed-up and automation of simple, tedious and repetitive service management
tasks currently performed most commonly by human operators. The ultimate re-
sult of this is potentially substantial cost savings to the operator. In particular, by
hiding low-level networking aspects that, especially in converged network scenar-
ios, can continuously change due to end users mobility, the LS/ASAM middleware
provides transparent service access in heterogeneous networks and becomes an
essential complement to (bearer unaware) service delivery platforms.
However, to achieve the potential of autonomic management systems in to-
days’ networks is not a straightforward task. Migrating intelligence and complex
management functions toward the edge of the network reduces the degree of man-
ual intervention needed, but increases somehow the complexity of the management
system itself. The network has indeed to be adaptable, but at the same time sta-
ble and controllable. Therefore, populating the networking environment with auto-
nomic software components requires some additional configuration and monitoring
capabilities. In this sense, middleware technologies for highly dynamic and hetero-
geneous networks must become able to monitor and control the middleware itself
by integrating with traditional, relatively static infrastructures often populated
by legacy solutions and adapting to different operating systems and connection
technologies. This is a challenging task that still requires additional investigation.
The system described in this paper has been implemented as a fully-functional
prototype with an accompanying scenario simulator for experimental evaluation.
The results presented in this paper are rather preliminary in that we cannot yet
report on the fully mediated solution, but they are nevertheless extremely en-
couraging. In particular, the demonstrable performance improvement with the
Autonomic Service Access Management 25
complete LS/ASAM suite in operation, as shown in Figure 7, is a quite significant
result.
Our ongoing and future work includes more refined and extensive character-
ization of LS/ASAM performance, especially on the network side, when adopting
different user and operators policies, network allocation strategies and algorithms.
While the LS/CA has been already successfully deployed in a variety of real-world
scenarios, the adoption of the LS/SAM requires some additional work given the
wide assortment of existing and upcoming service and network management archi-
tectures. In particular, by simulating and analyzing the LS/ASAM Suite perfor-
mance in a variety of networking scenarios and consequently refining the behavior
of the various system components, we expect to better characterize and conse-
quently improve performance and scalability. In addition, by means of selected
testbed demonstrations and experiments we are assessing the feasibility and com-
plexity of integrating LS/ASAM entities in specific service delivery frameworks
including IMS.
Acknowledgment
Many thanks to the colleagues at Whitestein Technologies who contributed sig-
nificantly toward this work, in particular Thomas Lozza, Martin Stangel, Oliver
Hoeffleur, Oliver Carl and the LS/CA team.
References
[1] Strassner, J.: Autonomic Networking - Theory and Practice. In: Proceedings 9th
IFIP/IEEE International Symposium on Integrated Network Management, Nice,
France (May 2005)
[2] Jennings, N.R.: Agent-Oriented Software Engineering. Lecture Notes in Computer
Science 1647 (1999) 1–7
[3] Ferrus, R., Gelonch, A., Sallent, O., Perez-Romero, J.: Vertical Handover Support
in Coordinated Heterogeneous Radio Access Networks. In: Proceedings 14th IST
Mobile and Wireless Communications Summit, Dresden, Germany (June 2005)
[4] Perkins, C.: RFC 3220: IP Mobility Support for IPv4 (January 2002)
[5] Kent, S., Atkinson, R.: RFC 2401: Security Architecture for the Internet Protocol
(November 1998)
[6] Yokoo, M., Hirayama, K.: Algorithms for Distributed Constraint Satisfaction: A
Review. Autonomous Agents and Multi-Agent Systems 3(2) (2000) 185–207
[7] Foundation for Intelligent Physical Agents: FIPA Iterated Contract Net Interaction
Protocol Specification (2001)
[8] Hofbauer, J., Sandholm, W.H.: On the global convergence of stochastic fictitious
play. Econometrica (70) (2002) 2265–94
[9] Shamma, J.S., Arslan, G.: Unified convergence proofs of continuous-time fictitious
play. IEEE Trans. on Automatic Control 49(7) (2004) 1137–42
26 M. Calisti, R. Ghizzioli and D. Greenwood
[10] Ahmavaara, K., Haverinen, H., Pichna, R.: Integration of wireless LAN and 3G wire-
less - Interworking Architecture between 3GPP and WLAN systems. IEEE Commu-
nications Magazine 11(41) (2003) 74–81
[11] Wang, X.G., Min, G., Mellor, J.E., Al-Begain, K., Guan, L.: An adaptive QoS frame-
work for integrated cellular and WLAN networks. Computer Networks 47(2) (2005)
[12] Song, W., Jiang, H., Zhuang, W., Shen, X.: Resource management for QoS support
in cellular/WLAN interworking. IEEE Network 19(5) (2005)
[13] 3GPP: TS 23.107 v7.4.0: Quality of Service (QoS) concept and architecture. (June
2006)
[14] Cuevas, M.: Admission control and resource reservation for session-based applica-
tions in next generation networks. BT Technology Journal 23(02) (2005)
[15] Kolbehdari, M., Lizotte, D., Shires, G., Trevor, S.: Session Initiation Protocol (SIP)
Evolution in Converged Communications. Intel Technology Journal 10(01) (2006)
[16] ABIresearch: IP Multimedia Subsystem Industry Survey Results (2005)
Monique Calisti, Roberto Ghizzioli, Dominic Greenwood
Whitestein Technologies AG
Pestalozzistrasse 24
CH-8032, Zurich,
Switzerland
e-mail: {mca,rgh,dgr}@whitestein.com
