Conference PaperPDF Available

A generic optimized time management algorithms (OTMA), framework for simulating large-scale overlay networks.

Authors:

Abstract and Figures

Recent evolutions in wireless networks will require more efficient use of the underlying parallel discrete-event simulation (PDES) synchronization protocols to accommodate the demand for large-scale network simulation. In this dissertation, we investigate underlying synchronization protocols to improve the performance of large-scale network simulators operating over PDES systems. We begin by proposing a generic optimized time management algorithms (OTMA) framework that combines the improved forms of synchronization protocols on a single platform. Particularly, for the proposed OTMA framework, we use the layered architecture approach to combine the optimized forms of conservative and optimistic time management algorithms. To support the implementation of the OTMA framework, a new m -LP (logical process) simulation model is proposed. One of the other challenges of large-scale network simulations is the lack of a realistic analytical and mathematical model for underlying PDES protocols. In this research work, the proposed OTMA framework integrates both conservative and optimistic synchronization algorithms on a single platform. In particular, for the purposes of this research, we provide an improved form of NMA by developing a new deterministic model that quantifies the performance dependent critical parameters for PDES systems. In addition, for the implementation of NMA, a new m -LP simulation model along with the varying parameters network topology is proposed. Finally, we provide a quantitative model to support the simulation results and experimental verifications for NMA. The current DES based simulators have a large end-to-end latency and poor memory utilization. OTMA framework will provide an improved form of the existing Time Wrap algorithm by proposing a new unacknowledged message list (UML) scheme. The proposed UML scheme will provide global synchronization among large number of nodes along with a fool proof solution for message transient and simultaneous reporting problems. To illustrate the implementation of the proposed UML scheme, two algorithms are proposed for coordinating and non-coordinating LPs. In order to further improve the global virtual time (GVT) computation process, synchronous barriers (such as tree and butterfly barriers) will be combined with the asynchronous algorithms (such as Time Wrap algorithm) to provide an efficient GVT computation mechanism for large-scale distributed networks.
Content may be subject to copyright.
A Generic Optimized Time Management Algorithms (OTMA) Framework for
Simulating Large-Scale Overlay Networks
Syed S. Rizvi and Khaled M. Elleithy
Department of Computer Science, University of Bridgeport, Bridgeport, CT 06604, USA
{srizvi,elleithy}@bridgeport.edu
Keywords: Distributed algorithms, discrete-event
simulation, overlay networks, synchronization algorithms.
Abstract
Recent evolutions in wireless networks will require
more efficient use of the underlying parallel discrete-event
simulation (PDES) synchronization protocols to
accommodate the demand for large-scale network
simulation. In this paper, we investigate underlying
synchronization protocols to improve the performance of
large-scale network simulators operating over PDES
systems. We begin by proposing a generic optimized time
management algorithms (OTMA) framework that combines
the improved forms of synchronization protocols on a single
platform. Particularly, for the proposed OTMA framework,
we use the layered architecture approach to combine the
optimized forms of conservative and optimistic time
management algorithms. To support the implementation of
the OTMA framework, a new m-LP (logical process)
simulation model is proposed along with the varying
parameters network topology that can show the
implementation of different components of discrete-event
simulation (DES).
1. INTRODUCTION
Next generation networks typically consist of heterogeneous
access networks with a high mobility rate of user terminals.
A foreseen future of these networks could be the support of
overlay networks by dedicated devices in the access
networks to alleviate user terminals from overlay traffic.
There has been much interest in emerging overlay networks
because they provide a good substrate for creating large-
scale data sharing, content distribution and application-level
multicast applications. Existing distributed DES systems do
not support simulation of networks with these special
properties. Conventional DES systems offer only limited
support to collect statistics and has a very simplified
underlying network layer without consideration of
bandwidth and latency costs. This makes it difficult to
simulate heterogeneous access networks and terminal
mobility. Most of these DES systems use either optimistic or
conservative algorithms as an underlying protocol for
achieving synchronization among huge number of nodes.
This much needed synchronization results in high end-to-end
delay, slower processing speed, large memory retirements,
and an excessive amount of transmission overhead.
Selection of nearby peers, redundant storage, efficient
search/location of data items, data permanence or
guarantees, hierarchical naming, trust and authentication,
self-organizing, massively scalable, and load balancing are
few of the many reasons for growth in the research and
development of overlay networks [1]. As a result, the
research and the developer community on distributed
systems, and in particular on overlay networks, needs tools
and frameworks for evaluating and developing their own
protocols and services, as well as against other protocols
with the same preconditions [2]. Packet-level simulators
(e.g., NS-3 [3], OMNET++ [4]) and high-level simulators
(e.g., p2psim [5], PeerSim [6], FreePastry [7], PlanetSim
[8]) are example of some of the existing DES based
simulators and frameworks.
As the demand for developing the new protocols and
services for overlay networks and peer-to-peer applications
increases, the necessity for an efficient distributed DES
framework for modeling and simulation is becoming
apparent. Several challenges occur with designing an
efficient DES framework for large-scale networks.
Firstly, traditional, packet-level DES simulators have a
true high cost in time and computer resources [2]. Currently,
DES simulators like NS-3 and OMNET++ do not scale to
thousands of nodes in a large overlay network [9]. Secondly,
existing high-level DES simulators such as p2psim [5],
PeerSim [6], FreePastry [7], and PlanetSim [8] provide
better performance and enable big scale network evaluations
but they do not support application level extension which is
an essential element in the development and deployment of
new protocols and services for large-scale networks. As a
result, developing and simulating new overlay services and
protocols, using the above mentioned DES simulators is not
possible. Thirdly, as far as we know, there is no common
DES framework available that combines the optimized
forms of the underlying synchronization protocols.
There are several reasons for this. In most DES
simulators for large-scale networks, optimization of the
27
underlying protocols is not given a first class status. Both
packet-level and high-level simulators use the TMA as an
underlying synchronization protocol in its original form.
Generally at the authoring stage, synchronization protocols
such as conservative NMA and optimistic Time Wrap
algorithms are imported in and complex functionality is
added by external scripts or programs. This may not only be
because of the absence of a common DES framework but
also because on the lack of analytical and quantitative
models present for these algorithms. Therefore, designing a
generic synchronization framework that combines the
optimized form of the underlying algorithms such as NMA
and Time Wrap is challenging. Finally, optimizing the
existing TMA requires the development of analytical and
quantitative models for each protocol is also an open
research problem.
In addition to the above mentioned challenges, building
a unified DES framework that can combine the optimized
forms of conservative and optimistic algorithms on a single
generic framework and developing their analytical and
quantitative models is not possible unless there is consensus
between network/protocol/services developers on a common
platform and technology. This problem does not exist in
traditional DES systems that model relatively small-scale
networks by using an un-optimized form of a
synchronization algorithm in its original form.
2. LARGE-SCALE NETWORK SIMULATORS
Several algorithms on general purpose sequential DES
systems and its variants were studied. The most popular
object-oriented discrete-event network simulator is NS-3 [9].
It is a packet-level simulator to simulate network protocols,
such as TCP, routing and multicast, on small networks.
However, the design of ns is such that simulation of large-
scale networks is difficult, if not impossible, due to
excessive memory and CPU time requirements. J-Sim [10]
(formerly called JavaSim) is another component-based
network simulation framework which is similar to NS-3.
OPNET [11] modeler is the leading commercial network
simulator for studying and evaluating networks and
distributed systems. OMNeT++ [4] is an open-source
simulation environment that is similar to OPNET. All of the
above mentioned general purpose sequential DES systems
can simulate small networks. However, they cannot scale to
networks with more than a few hundreds of nodes.
A number of PDES systems were identified that use
TMA to provide synchronization among participating
processors that perform concurrent execution of discrete-
events in a distributed network. Reference [12] proposed
PDNS (parallel distributed network simulator) which is an
enhancement to NS-3 to utilize parallel computation on
multiple machines. PDNS uses a conservative TMA for
synchronization. The main objective of PDNS is to reduce
the memory requirement by distributing the simulation load,
and thus improve the overall latency. The design of both NS-
3 and its successor PDNS, using the hybrid of Tcl and C
languages, leads to substantial memory consumption in
many cases [13].
The Georgia Tech Network Simulator (GTNetS) [13] is
a component-based network simulator aimed to provide a
structured simulation environment. GTNetS uses
conservative TMA as an underlying synchronization
algorithm. GTNetS uses many techniques used in PDNS to
enable parallel computation. SSFNet [14] is a scalable high
performance simulation platform which is designed to
operate on a share–memory multiprocessor, with global state
available to all simulation threads. It provides a unified
interface for DES. SSF also provides a high-level modeling
language to describe modeling environments. Reference [15]
proposed a scalable simulation framework (GloMoSim) [16]
to simulate wireless and wired network systems. The
GloMoSim applies the PDES functionality provided by
PARSEC [15]. GloMoSim uses a layered model similar to
the OSI layers. PARSEC [15] is a DES language that adopts
the process interaction approach to DES. In PARSEC, an
object in the physical system is represented by a LP.
Interactions among physical processes (events) are modeled
by exchanging time-stamped event messages between the
corresponding LPs.
Several DES based simulators designed specifically for
overlay systems that have an underlying conservative or
optimistic TMA were studied. P2PSim [5] is an open-source
discrete-event simulator developed for P2P system
simulations. It supports several existing P2P protocols:
Chord [17], Tapestry [18] etc. For underlying network
topologies, P2PSim supports a number of topology models
including constant distance topology, DV graph, end-to-end
time graph, Euclidean graph, G2 graph, etc. However, it
does not support distributed simulation. PeerSim [6] is a
Java-based P2P simulator that can be used to simulate both
structured and unstructured P2P protocols. PeerSim provides
two simulation models: cycle-based model and event-driven
model. PeerSim uses a simplified Internet model of the
message passing and it can simulate networks with large
sizes in the cycle based model (up to 1 million nodes).
However, PeerSim also does not support distributed
simulation and parallel execution of events. Overlay Weaver
[19] is designed as a simulation tool for easy development
and testing of overlay protocols and services. Overlay
Weaver implements common APIs such as DHT and
multicast so that users can study and evaluate services built
on top of these APIs. It implements Pastry [7], Chord [17],
and Tapestry [18], etc. Overlay Weaver has a modular
architecture. The routing layer is composed of several
components such as routing algorithm, message service, etc.
Although, the Overlay Weaver is scalable to large size
networks, it also does not support parallel execution of
events.
Lin, Pan, Guo, and Zhang [20] proposed a distributed
simulation architecture for large-scale overlay networks that
typically involves more than 1 million nodes. Their proposed
architecture uses conservative synchronization as an
28
underlying TMA. To avoid the Lookahead problem of
conservative algorithm and maximize the execution speed,
they introduced a slow message relaxation scheme that
handles straggler events that may arrive later in the
simulation time due to network delays. Although, their
proposed scheme may achieve the peak speedup for large-
scale networks, the statistical accuracy of the distributed
simulation may compromise due to the occurrence of late
messages (i.e., slow messages) [10]. Late messages are
referred as transient messages that may result in an
inaccurate GVT computation. As a result, it is hard to assure
the statistical accuracy of the final simulation results
produce by the simulation engine.
As discussed above, the current DES systems (both
sequential and parallel) are mostly focused on providing a
simulation framework to study the network protocols and
services. However, none of them are perfectly suitable for
simulations of overlay networks in large-scale. The current
overlay simulators are not scalable in simulating large
overlay networks since they do not support distributed
simulation such as P2PSim [5], PeerSim [6], Overlay
Weaver [19] etc.
3. OPTIMIZED TIME MANAGEMENT
ALGORITHM (OTMA) FRAMEWORK
System architecture for a generic OTMA framework is
proposed in Fig.1 that combines different functionalities of
PDES simulator in a three layer model. A complete cycle of
modeling and simulation is decomposed into the proposed
three layers architecture. In the next section, internal
architecture of the proposed OTMA framework for PDES
system is presented.
3.1. OTMA Generic System Architecture
The proposed OTMA generic system architecture is
presented in Fig. 1. For OTMA framework, we adopt a top
to bottom layered architecture style to separate the distinct
functionalities that are needed for modeling a large-scale
Overlay network and executing the simulation protocols.
The OTMA framework consists of the following three
layers: Model Abstraction Layer (MAL), DES Optimized
Protocol Layer (OPL), and System Implementation Layer
(SIL).
The top layer, MAL, provides an interface between a
simulation designer/engineer and the OTMA framework.
There are two main sub-layers of MAL: (i) DES interface
and (ii) Model formalization layer (MFL). DES interface
allows a simulation designer to interact with the layers of
OTMA framework in order to represent and execute model
components and simulation protocols in a distributed
system.
For instance, a simulation designer can use the DES
interface to use an abstract simulator for specifying the
components of the physical system and formally define the
model. The bottom sub-layer of MAL is a model
formulization layer (MFL) that allows a simulation designer
to provide an abstract view of the model under
consideration. Model formulization layer can adopt any third
party abstract simulator for this task. We believe that the
first layer is well researched and most of the existing third
party software and tools can be employed at sub layers of
MAL. Therefore, for this research work, we focused on the
middle and the bottom layers of the OTMA framework.
The middle layer, DES OPL, provides an optimized
form of TMA that can be used to provide synchronization
among the participating LPs. In particular, DES OPL
consists of distributed simulation protocols (e.g., NMA or
Time Wrap algorithms) that provide services (such as
ordering of event messages, transmission of synchronization
messages, maintaining the virtual time etc.) that are
necessary for different components of the DES systems
(such as inter LP communication) to correctly interoperate.
For this particular layer of OTMA framework, our main
contribution will be the development of an optimized form
of Time Wrap algorithm. The main significance of this
development is that it provides a new UML scheme that not
only solves the well known transient message problem but
also optimizes the performance by reducing the latency and
memory requirements. The detail of UML scheme is
presented in [22]. In addition, this layer provides a
deterministic model for NMA that allows the simulation
designer to choose one of the most appropriate DES
protocols from OPL with respect to the model specified at
MAL of OTMA framework. The details of the proposed
deterministic model can be found in [23].
The bottom layer, SIL, of OTMA framework provides
the implementation platform where the system that specified
at the model abstraction layer (MAL) is implemented. In
particular, for this layer, we propose a new internal
architecture of an m-LP simulation model where m
represents the total number of participating LPs in DES
system. Our proposed m-LP simulation model will utilize
one of the optimized protocols from the above layer (i.e.,
Fig.1. OTMA framework layered architecture and services
29
DES OPL) to optimize the overall performance of PDES
system. A two dimensional clock (TDC) is proposed for this
layer that controls the simulation/virtual time. The bottom
layer is mainly responsible for all inter-LP communication,
event messages execution, and event and clock management.
Our second contribution for this layer is the design of a new
LP controller that consists of three main sub components: (i)
Simulation Executive (SE), (ii) Simulation Application (SA),
and (iii) Inter LP Communication Interface (CI). The detail
of these components will be provided subsequently.
3.2. Bottom Layer - System Implementation Layer (SIL)
The system component view of OTMA framework was
proposed in Fig. 1. As mentioned before, in the proposed
OTMA architecture, SIL is mainly responsible for all inter
LP communication, LP mapping, local and remote event
message management/execution, TDC implementation, and
LP synchronization. To address all these issues, we propose
a new m-LP simulation model, architecture for an internal
LP’s main controller, and TDC. For the sake of simplicity,
we first present the proposed internal architecture of a
generic simulation model that consists of m number of LPs
where each LP has i number of input lines and o number of
output lines. The internal architecture of an LP is presented
subsequently to show the different components such as
simulation executive and event pool. Finally, in this section
we present the architecture of LP’s main controller.
3.2.1 Proposed m-LP simulation model
The proposed internal architecture of a generic simulation
model for m number of LPs is shown in Fig. 2. It consists of
primarily three components:
1. Input Queues
2. Output Queues
3. Two Dimensional Clock (TDC)
The input and output queues carry all inter LP
communication in the proposed simulation model by sending
and receiving the event messages. For the proposed model,
we assume a mesh topology where each LP is connected to
the other LP via a direct communication link for maximizing
the internetworking and inter-LP communication. Each LP,
therefore, maintains i number of input lines to receive the
event messages from input neighbors and o number of
output lines for scheduling the event messages for other
neighboring LPs. For an m-LP simulation model, i input
lines = o output lines = m-1 neighboring LPs. Input and
output queues act as a communication channel for
exchanging event messages to execute and progress the
simulation as well as send synchronization messages to
synchronize the participating LPs. We assume that each LP
maintains a TDC which is mainly responsible to ensure that
the local causality requirement must not be violated. In
particular, TDC maintains two clock times: one for each of
its input and output neighbors as shown in Fig. 3.
The first clock time is the minimum receiving time
(MRT) for the input neighbors LPs whereas the second clock
time is the minimum sending time (MST) for the output
neighbors LPs. The MRT indicates an earliest time when an
Fig.2. m-LP Simulation model: m number of LPs with I number of input queues
and O number of output queues per LP
Fig.3. TDC implementation on m-LP simulation model with minimum
receiving time (MRT) and minimum sending time (MST)
30
LP may receive an event message from one of its input
neighboring LPs, where as the MST represents an earliest
time by which an LP can send an event message to one of its
neighboring LPs. TDC updates and adjusts the MST with
respect to the current simulation time (Ts) of that LP whereas
the MRT is updated with respect to the smallest value of
LBTS.
This can also be expressed as:
1 2
,,......, M
MRT Min LBTS LBTS LBTS
= for m-LP
simulation model where LBTS1 corresponds to LP1, LBTS2
corresponds to LP2 and LBTSm corresponds to LPm. Each LP
computes its own LBTS value to synchronize itself with the
other LP by exchanging this information. LBTS value for an
LPi is computed as follows:
3.2.2 Proposed internal architecture of an LP
An internal architecture of an LP is shown in Fig. 4. In m-LP
simulation model, an LP is connected to other neighboring
LPs via a direct input communication link to receive remote
event messages scheduled by the other neighboring LPs.
Similarly, the execution of an event message may schedule a
new event message for one of the neighboring LPs. This
event message is referred as a remote event message and it
will be sent to the destination LP using one of the direct
output lines as shown in Fig. 4. Each LP in the proposed
architecture implements the TDC that maintains the MRT for
the input lines and MST for the output lines to ensure that the
LP must process the event messages strictly in the non-
decreasing time-stamp order and thus avoids the violation of
causality constraint requirement. The internal architecture of
the main controller of Fig. 4 will be elaborated subsequently.
3.2.3 Proposed architecture for LP’s main controller
The proposed internal structure of a main controller of an LP
is shown in Fig. 5. Each LP is mapped to one of the
processors and encapsulates the controller that consists of 3
components as mentioned below.
1. Simulation Executive (SE)
2. Simulation Application (SA)
3. Inter LP Communication Interface (CI)
The Simulation Executive (SE) is completely
independent to the model of the physical system and
therefore can be used to simulate several different types of
systems. SE consists of two main components: event
scheduler (ES) and TDC. The main responsibility of SE is to
ensure reliable exchange of remote event messages among m
number of LPs as well as periodic exchange of
synchronization messages between the participating LPs.
In our proposed architecture, SE is the only sub
component that can communicate with the other
participating LPs via the CI. In other words, all inter LP
communication (i.e., the exchange of both event and
Fig.4. Internal architecture of an LP with I number of input lines and O
number of output lines per LP
Fig.5. An illustration of internal structure of a main controller of an LP with the
exchange of event-messages and procedure calls
31
synchronization messages) has to be done using the
corresponding SE and CI.
For the proposed controller architecture, we use FIFO
queues with a static mesh communication topology similar
to what is assumed in [21]. ES maintains one FIFO queue
for each of the input neighbors where all the event messages
are stored according to their time-stamps. Since we are using
FIFO queues for the proposed architecture, the head of each
queue will have the event message that has the smallest
time-stamp within that queue. The use of FIFO queue,
therefore, avoids the use of out of order sequencing that may
be required to select the event message with the smallest
time-stamp for execution.
In addition, the use of TDC ensures that the event
messages are stored in each FIFO queue of an LP in strictly
non decreasing time-stamp order. Each time when an LP
receives a remote event message from one of the
neighboring LPs, TDC compares the MRT to the time-stamp
of the received message to verify whether the time-stamp of
the received message is smaller or larger than MRT. If time-
stamp of the received event message is equal or greater than
the MRT, the event message will be accepted and stored in
the corresponding FIFO queue else it will be rejected since
this will violate the causality constraint requirement. ES
repeatedly removes the event containing the smallest time-
stamp from the FIFO queues, call the TDC to advance the
current simulation time (Ts) and update the MST and MRT
accordingly, and finally call the SA for event execution. The
two sub components of SE are shown in Fig. 6.
Simulation Application (SA) consists of two main
components: Event Handler (EH) and local state variables.
When an SA receives the execution call from SE, it calls the
EH that processes the event. The processing of an event
message may result two things: (i) the scheduling of new
event messages for the future simulation time, and (ii)
change the state of the physical system (modify the local
state variables to reflect changes in the state of the physical
system).
EH has the capability to determine whether the newly
generated event message(s) is local or remote. If the
preprocessing of an event message results in the scheduling
of one or more remote event messages, EH will call the SE
which in turns send the remote event messages to CI.
However, if EH determines that the newly generated event
messages are the local event messages, it calls ES directly to
schedule local events. If the execution of an event message
changes the state of the physical system, that change in the
state must be reflected by updating/modifying the local state
variables. The interaction of two sub-components of SA with
SE is shown in Fig. 7.
In our proposed architecture, SE is directly connected to
Communication Interface (CI) to send and receive time-
stamped event messages as well as synchronization
messages. Each LP has its own CI which is connected to the
Fig.7. Interaction between simulation application (SA) a nd simulation
executive (SE) to schedule and execute local and remote event-messages
Fig.6. Internal structure of simulation executive (SE) with the two main sub-
components (TDC and ES) interacting with SA and CI
32
other CIs of the neighboring LPs. In the context of NMA,
whenever TDC updates the current simulation time after the
retrieval of an event message from FIFO queue, a
synchronization message will be sent by SE to other LPs
using the respective CI.
The synchronization message (also referred as null
message) will contain the time-stamp equal to MST of the
sending LP. When CI of an LP receives an outgoing remote
event message from the SE, it checks the destination and
sends the message using the appropriate output line. When a
CI receives an incoming remote event message from the
other CI of an LP, it sends the message to the SE which in
turns sends that event message to ES. This implies that EH
directly calls SE for outgoing remote event messages
whereas SE directly calls ES for incoming remote event
messages. In addition, EH can directly call ES to schedule
local event messages.
3.3 Middle Layer DES Optimized Protocol Layer
(OPL)
DES OPL is a middle layer in the proposed OTMA
framework architecture that provides optimized form of
TMA such as conservative NMA and optimistic Time Wrap
algorithm. When a simulation designer is done with the
formal specification of the target system by developing an
abstract model, he/she can choose the PDES simulation
protocol from the lower layer (i.e., DES OPL) by analyzing
that which one of the TMAs is most appropriate with respect
to the abstract view of the model presented at the top layer
(i.e., MAL). The internal architecture of DES OPL layer can
be seen in Fig. 8.
For this particular layer of OTMA framework, our main
contribution will be the development of an optimized form
of Time Wrap algorithm. The detail of the optimized form of
Time Wrap algorithm is presented in [22]. In addition, this
layer provides a deterministic model for NMA that allows
the simulation designer to choose one of the most
appropriate DES protocols from OPL with respect to the
model specified at MAL of OTMA framework. The details
of the proposed deterministic model can be found in [23].
4. CONCLUSION
In this paper, we presented a generic OTMA framework that
combines the optimized forms of synchronization protocols
on a single platform. Specifically, a layered architecture
approach was proposed to combine the optimized forms of
conservative/optimistic algorithms on a generic
synchronization framework. To support the implementation
of the proposed OTMA framework, a new m-LP simulation
model was proposed. We adopted a layered architecture
style for the proposed OTMA framework that allows the
simulation designers to further contribute in each
independent layer of the framework. In addition, our
analysis showed that the proposed OTMA framework is
protocol independent and modular in its architecture that
results in a strong simulation model and consequently better
simulation results. Our proposed OTMA framework can be
used as commonly agreed reference architecture to provide a
common platform for all TMA along with the other
functionalities (such as DES interface and inter LP
communications) that are needed for modeling and
simulating a large-scale PDES system. Barring legal
ramifications, OTMA framework can be used by the
simulation engineers to develop a network simulator for
modeling and simulating large-scale networks at the
minimum cost (i.e., reduced transmission overhead, latency,
processor idle time, and memory requirements). A direct
application of OTMA framework would be to model and
simulate a physical system in a rather simplified and
efficient way using the three layers architecture.
Alternatively, the middle and the bottom layers of OTMA
framework can be adopted in any general purpose sequential
and parallel DES based network simulators to provide
effective synchronization among participating LPs. For
future work, we will integrate other TMA in the DES OPL
layer of the proposed OTMA framework to extend the
synchronization platform for PDES systems.
REFERENCES
[1] K. Aguilera, A. Merchant, M. Shah, A. Veitch, and C. Karamanolis,
“Sinfonia: A new paradigm for building scalable distributed systems,”
ACM Trans., Computer Systems, vol. 27, No. 3, Article 5, November
2009.
[2] J. P. Ahulló, P. G. López, M. S. Artigas, and M. A. Arias, “An
extensible simulation tool for overlay networks and services,”
Proceedings of the 2009 ACM Symposium on Applied Computing,
2009, pp. 2072-2076.
[3] The NS-3 Network Simulator. Available: http://www.nsnam.org/
[4] OMNET++. Available: http://www.omnetpp.org/
Fig.8. Interaction of DES-OPL layer with Model Abstraction Layer (MAL)
and System Implementation Layer (SIL) of OTMA framework
33
[5] p2psim. Available: http://pdos.csail.mit.edu/p2psim/.
[6] M. Jelasity, A. Montresor, G. Jesi, and S. Voulgaris. The Peersim
Simulator. Available: http://peersim.sf.net.
[7] A. Rowstron and P. Druschel, “Pa stry: Scalable, decentralized object
location, and routing for large-scale peer-to-peer systems,” In the
Proceeding of IFIP/ACM International Conference on Distribut ed
Systems Platforms, November 2001, vol. 2218, pp. 329–350.
[8] P. Garc´ıa, C. Pairot, R. Mond´ejar, J. Pujol, H. Tejedor, and R. Rallo,
“PlanetSim: A new overlay network simulation framework,” Software
Engineering and Middleware, 2005, pp. 123-136.
[9] L. Cheng, “Simulation and topology generation for large-scale
distributed systems,” Ph.D. dissertation, University of British
Columbia, September 2009.
[10] J. Liu, “Parallel discrete-event simulation,” Ph.D. dissertation, School
of Computing and I nformation Sciences, Florida International
University, February 2009.
[11] OPNET. Available: http://www.opnet.com.
[12] S. L. Ferenci, K. S. Perumalla, and R. M. Fujimoto, “An approach for
federating parallel simulators,” In the Proceedings of the 14th
Workshop on Parallel and Distributed Simulation, Washington, DC,
USA, 2000, pp. 63 – 70.
[13] G. F. Riley, “The Georgia tech network simulator,” In the Proceedings
of the ACM Sigcomm Workshop on Models, Methods and Tools for
Reproducible Network Research, N ew Y ork, NY, USA, 2003, pp. 5-
12.
[14] S. Yoon and Y. B. Kim, “A design of network simulation environment
using SSFNet,” First International Conference on Advances in S ystem
Simulation, 2009, pp.73-78.
[15] M. Takai, Y. A. Chen, X. Zeng, and J. Martin, “Parsec: A parallel
simulation environment for complex systems,” IEEE Computer, pp.
31 - 77, 1998.
[16] X. Zeng, R. Bagrodia, and M. G. Glomosim, “A librar y for parallel
simulation of large-scale wireless networks,” In the Proceedings of
Workshop on Parallel and Distributed Simulation, 1998, pp. 154-161.
[17] I. Stoica, R. Morris, D. Karger, F. Kaashoek, and H. Balakrishnan,
“Chord: A scalable peer-to-peer lookup service for internet
applications,” In the Proceedings of the 2001 ACM SIGCOMM
Conference, 2001, pp. 149 - 160.
[18] B. Zhao, L. Huang, S. Rhea, J. Stribling, A. Joseph, and J. D.
Kubiatowicz, “Tapestry: A global-scale overlay for rapid service
deployment,” IEEE J-SAC, vol. 22, No. 1, pp. 41 - 53, January 2004.
[19] K. Shudo, Y. Tanaka, and S. Sekiguchi, “Overlay weaver: An overlay
construction toolkit,” Computer Communications (Special Issue on
Foundations of Peer-to-Peer Computing), vol. 31, pp. 402 - 412, 2007.
[20] S. Lin, A. Pan, R. Guo, and Z. Zhang, “Simulating large-scale P2P
systems with the WiDS toolkit,” In the Proceedings of the 13th IEEE
International Symposium on Modeling, Analysis and Simulation of
Computer and Telecommunication Systems, 2005, pp. 415 – 424.
[21] M. Chetlur and P. Wilsey, “Local fossil identification in time warp
simulations,” International Journal of Simulation and Process
Modelling, vol. 5, No. 2, pp. 95 – 103, 2009.
[22] Syed S. Rizvi, Aasia Riasat, and Khaled M. Elleithy. An Efficient
Optimistic Time Management Algorithm for Discrete-Event
Simulation System, International Journal of Simulation Modeling,
Vol. 9. No. 3, pp. 117 - 130, September 2010.
[23] Syed S. Rizvi, Aasia Riasat, and Khaled Elleithy, “Reducing Null
Message Traffic in Large Parallel and Distributed Systems,” 13th IEEE
Symposium on Computers and Communications (ISCC'08), pp. 1115 –
1121, July 6 - 9, 2008, Marrakech, Morocco.
34
... Some recent research on conservative algorithms in DES can be found in [11][12][13][14]. An effort to combine conservative and optimistic synchronization algorithms on a common layered architecture framework is proposed in [15]. ...
Article
Full-text available
This paper presents a new logical process (LP) simulation model for distributed simulation systems where Null Message Algorithm (NMA) is used as an underlying time management algorithm (TMA) to provide synchronization among LPs. To extend the proposed simulation model for n number of LPs, this paper provides a detailed overview of the internal architecture of each LP and its coordination with the other LPs through sub-system components and models such as communication interface and simulation executive. The proposed architecture of LP simulation model describes the proper sequence of coordination that need to be done among LPs though different subsystem components and models to achieve synchronization. To execute the proposed LP simulation model for different set of parameters, a queuing network model is used. Experiments will be performed to verify the accuracy of the proposed simulation model using the pre-derived mathematical equations. Our numerical and simulation results can be used to observe the exchange of null messages and overhead indices.
Conference Paper
Cognitive Radio (CR) is a promising solution to solve the problem of limited spectrum and poor bandwidth utilization in wireless networks by opportunistically exploiting the available wireless spectrum. Despite the numerous advantages of CR technology, it imposes unique challenges for modeling and simulation due to dynamic spectrum sensing and the high fluctuations in the spectrum utilization as well as diverse set of components that form a CR network. Therefore, to develop a model for CR network and execute the distributed simulation, it is critical to have a complete understanding of (1) an analytical model that incorporates all components of a CR network and (2) a framework that layout all necessary tools and services needed for a successful modeling and simulation. In this paper, we have two research contributions. First, we present an analytical model of CR by identifying components necessary for a design of a typical CR and classify them in the context of modeling and simulation. A design of a CR network includes the software defined radio (SDR), spectrum sensing, and communication channels. The layers of the analytical model encapsulate the SDR, spectrum sensing techniques, and the communication channels. Second, we construct a theoretical representation of a CR by specifying its unique characteristics needed for modeling and simulation. We show the mapping of these characteristics into a layered architecture framework which ease the model formulization and simplify the simulation of a complex CR network.
Article
Full-text available
Time Wrap algorithm is a well-known mechanism of optimistic synchronization in a parallel discrete-event simulation (PDES) system. It offers a run time recovery mechanism that deals with the causality errors. For an efficient use of rollback, the global virtual time (GVT) computation is performed to reclaim the memory, commit the output, detect the termination, and handle the errors. This paper presents a new unacknowledged message list (UML) scheme for an efficient and accurate GVT computation. The proposed UML scheme is based on the assumption that certain variables are accessible by all processors. In addition to GVT computation, the proposed UML scheme provides an effective solution for both simultaneous reporting and transient message problems in the context of synchronous algorithm. To support the proposed UML approach, two algorithms are presented in details, with a proof of its correctness. Empirical evidence from an experimental study of the proposed UML scheme on PHOLD benchmark fully confirms the theoretical outcomes of this paper.
Article
Abstract Anumber,of library-based parallel ,and sequential network,simulators ,have ,been ,designed. This paper describes a library, called GloMoSim (for Global Mobile system Simulator), for parallel simulation of wireless networks. GloMoSim has been designed to be ,extensible and composable: the communication ,protocol stack for wireless networks is divided into a set of layers, each with its own API. Models of protocols at one layer interact with those at a lower (or higher) layer only via these APIs. The modular,implementation,enables consistent comparison,of multiple,protocols ,at a ,given ,layer. The parallel implementation,of GloMoSim ,can be executed ,using a variety of conservative synchronization protocols, which include,the ,null ,message ,and ,conditional ,event algorithms. This paper describes the GloMoSim library, addresses,a number ,of issues ,relevant ,to its parallelization, and presents a set of experimental results onthe IBM 9076 SP, a distributed memory multi- computer. These experiments use models constructed from the library modules. 1,Introduction The,rapid ,advancement ,in portable ,computing platforms and wireless communication,technology has led tosignificant interest in mobile ,computing ,and mobile networking. Two primary forms of mobile ,computing ,are becoming popular: first, mobile computers continue to heavily use wired network infrastructures.Instead of being hardwired to a single location (or IP address), a computer can,dynamically ,move ,to multiple ,locations ,while maintaining,application transparency. Protocols such as
This paper presents an enhanced process model in an optimistic distributed simulation to develop a method for tracking causality and safe-time and its application in the design of a local fossil identification technique. The fossil identification process is reduced to a local operation that operates without requiring a GVT estimation method. The information needed for fossil identification is captured during normal event processing from the enhanced process model. This technique assumes FIFO communication channels and simulation model with static communication topology. A proof of correctness of the fossil identification technique using this enhanced process model is presented.
Article
Network simulation tool is needed to verify the functionalities and performance of networks. The SSFNet (Scalable Simulation Framework Network Models) is a network simulation tool using open source software with various network simulation applications. It has been designed for the expansion of network including topology, protocols, traffic, and etc, and is able to support simulation for the large-scale network like Internet. However it is not easy for general users to perform network simulation using SSFNet because the SSFNet does not provide users with any supplementary tools for designing of network elements and topology, and analyzing of simulation results. The network modeling and analysis process must be done manually by users themselves. This circumstance makes it difficult to perform reliable network simulation. In this paper, we design a network simulation environment available for SSFNet. Using this environment, users can build network simulation model effectually and analyze the simulation results without difficulty.
Article
We introduce a new network simulation environment, developed by our research group, called the Georgia Tech Network Simula-tor (GTNetS). Our simulator is designed specifically to allow much larger–scale simulations than can easily be created by existing net-work simulation tools. The design of the simulator very closely matches the design of real network protocol stacks and hardware. Thus, anyone with a good understanding of networking in general can easily understand how the simulations are constructed. Further, our simulator is implemented completely in object–oriented C++, which leads to easy extension by users to experiment with new or modified behavior of existing simulation models. Our tool is de-signed from the beginning with scalability in mind, including the support for distributed simulations on a network of workstations as part of the basic design. We give an overview of the features of GTNetS, and present some preliminary scalability results we have obtained by running GTNetS on a computing cluster at the Pittsburgh Supercomputer Center.
Chapter
Parallel discrete-event simulation (PDES), sometimes simply referred to as parallel simulation, is concerned with the execution of discrete-event simulation on parallel computers. PDES has been recognized as a challenging field of research bridging modeling and simulation, and high-performance computing. By exploiting the potential parallelism in a simulation model, PDES can overcome the limitations imposed by sequential simulation, both in the execution time and the memory space. It has been demonstrated to be a viable technique for solving large-scale complex models. In this article, we provide a brief overview of the current state of PDES, identify its fundamental challenges, and discuss existing principal solutions resulting from three decades of intensive research in this field. Further, we report specific research advances in high-performance modeling and simulation of large-scale computer networks as an exemplar of typical PDES applications. Keywords: parallel simulation; PDES; conservative synchronization; optimistic synchronization; network simulation
Article
A layered model of structured overlays has been proposed and it enabled development of a routing layer independently of higher-level services such as DHT and multicast. The routing layer has to include other part than a routing algorithm, which is essential for routing. It is routing process, which is common to various routing algorithms and can be decoupled from a routing algorithm.We demonstrated the decomposition by implementing an overlay construction toolkit Overlay Weaver. It facilitates implementation of routing algorithms and we could multiple well-known algorithms just in hundreds of lines of code with the toolkit. The decomposition also enables multiple implementations of the common routing process. Two implementations the toolkit provides perform iterative and recursive routing, respectively. Additionally, to our knowledge, the toolkit is the first feasibility proof of the layered model by supporting multiple algorithms and the higher-level services.Such modular design contributes to our goal, which is facilitation of rapid development of realistic routing algorithms and their application. We demonstrates that Overlay Weaver supports the goal by conducting large-scale tests and comparisons of algorithms on a single computer. The resulting algorithm implementations work on a real TCP/IP network as it is.
Conference Paper
This paper presents the design and evaluation of Pastry, a scalable, distributed object location and routing substrate for wide-area peer-to-peer applications. Pastry performs application-level routing and object location in a potentially very large overlay network of nodes connected via the Internet. It can be used to support a variety of peer-to-peer applications, including global data storage, data sharing, group communication and naming. Each node in the Pastry network has a unique identifier (nodeId). When presented with a message and a key, a Pastry node efficiently routes the message to the node with a nodeId that is numerically closest to the key, among all currently live Pastry nodes. Each Pastry node keeps track of its immediate neighbors in the nodeId space, and notifies applications of new node arrivals, node failures and recoveries. Pastry takes into account network locality; it seeks to minimize the distance messages travel, according to a to scalar proximity metric like the number of IP routing hops. Pastry is completely decentralized, scalable, and self-organizing; it automatically adapts to the arrival, departure and failure of nodes. Experimental results obtained with a prototype implementation on an emulated network of up to 100,000 nodes confirm Pastry’s scalability and efficiency, its ability to self-organize and adapt to node failures, and its good network locality properties.