An Efficient Dynamic Load Balancing Scheme for Distributed Simulations on a Grid Infrastructure

Ajaltouni, Elie El; Boukerche, Azzedine; Zhang, Ming

doi:10.1109/ds-rt.2008.38

Cited by 29 publications

(10 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The simulation centred approaches observe the execution performance status based on simulations, as a method of indirectly evaluating the resources' load. In this case, such schemes measure and analyze the execution pace of simulation entities [7] [18]. Focusing on this last approach, the cited balancing designs present particular limitations, and observing that a centralized balancing scheme [19] and a distributed balancing scheme [1] have been developed.…”

Section: Related Workmentioning

confidence: 98%

Dynamic load redistribution based on migration latency analysis for distributed virtual simulations

Grande

Boukerche

2011

2011 IEEE International Workshop on Haptic Audio Visual Environments and Games

Self Cite

View full text Add to dashboard Cite

Distributed virtual simulations deployed on shared resources can frequently undergo loss of performance due to external background load, improper placement of simulation entities, or dynamic simulation load changes. The High Level Architecture (HLA) was designed as a solution for coordinating the execution of distributed simulations. Even though this framework offers management services to organize such simulations, it does not provide mechanisms for detecting and controlling load imbalances. Several balancing approaches have been designed aiming at a generic scheme for solving load imbalance issues of distributed simulations, but these approaches are concerned with issues of specific simulation applications or are unaware of environment characteristics. To overcome such limitations, a dynamic, distributed balancing scheme has been developed. However, the scheme is not aware of federate migration latencies. Since migration latency directly influences balancing efficiency and responsiveness, a redistribution scheme is proposed to measure migration delays and use such delays in the balancing algorithm to determine load deployment changes. These delays are used in a cost function that determines the redistribution behaviour of the balancing scheme. Experiments have been performed to analyze the performance gain of the proposed scheme when migration procedures introduce costly latencies into simulations.

show abstract

Section: Related Workmentioning

confidence: 98%

Dynamic load redistribution based on migration latency analysis for distributed virtual simulations

Grande

Boukerche

2011

2011 IEEE International Workshop on Haptic Audio Visual Environments and Games

Self Cite

View full text Add to dashboard Cite

show abstract

“…al. [8] works in alternate computation and communication balancing cycles by collecting CPU consumption and communication load of each simulation entity from JXTA grid-based [18] simulations. In these approaches, CPU load provides an applicationindependent metric to detect load imbalances, but the collected CPU regards only the simulation entities, neglecting external load and heterogeneity of resources.…”

Section: Related Workmentioning

confidence: 99%

Dynamic Load Balancing Using Grid Services for HLA-Based Simulations on Large-Scale Distributed Systems

Boukerche

Grande

2009

2009 13th IEEE/ACM International Symposium on Distributed Simulation and Real Time Applications

Self Cite

View full text Add to dashboard Cite

HLA-based simulations, as any distributed computing application, can undergo critical performance issues due to load imbalances on large-scale, heterogeneous, non-dedicated distributed systems. Such imbalances are produced by HLA simulation entities that can dynamically change their computation and communication load during their execution time, so an initial static load deployment is incapable of providing simulations complete and even distributed resources usage. Moreover, because the computing resources are non-dedicated, unknown external applications can generate load for any computing resource, increasing the imbalances' unpredictability. Thus, in order to re-allocate resources for an HLA simulation during its execution time, an hierarchical dynamic load balancing system is introduced. The system manages a simulation's workload by monitoring the distributed load through the MDS Grids' service; by identifying load imbalances according to a load sharing policy; by re-allocating resources according to defined policies; and by migrating federates through the GRAM Grids' service, a migration proxy, and peer-to-peer state transfer. By keeping the load evenly partitioned on the distributed system, such a devised system successfully improved the simulations' performance. The experimental results and comparative analyses between balanced and non-balanced simulations proved the efficiency of the proposed dynamic load balancing system.

show abstract

“…Global or partial information policies can be defined in a load balancing algorithm to decide the transfer of load among processors. With a global policy, load balancing algorithms require the knowledge about the load state of all the processors in the system [1], [2]. Clearly, to obtain this information when the number of processors increases, a large amount of communications needs to be done generating possible bottlenecks (showing a lack of scalability).…”

Section: Introductionmentioning

confidence: 99%

Load Balancing Algorithms with Partial Information Management for the DLML Library

Santana-Santana¹,

Castro-García²,

Aguilar-Cornejo³

et al. 2010

2010 18th Euromicro Conference on Parallel, Distributed and Network-Based Processing

View full text Add to dashboard Cite

Load balancing algorithms are an essential component of parallel computing reducing the response time of applications. Frequently, balancing algorithms have a centralized behavior requiring a lot of messages to operate, thus causing scalability problems. A solution to improve scalability is to define a decentralized algorithm, avoiding the generation of bottlenecks. DLML (Data List Management Library) is a tool that, in a transparent way, allows the parallel processing of data that are organized through a List. One drawback of this tool is the global bidding algorithm used to distribute the data (work) generated during the execution. In this paper two load balancing algorithms for DLML handling partial information are proposed. The first algorithm considers a logical Torus topology and the second one follows a Binary Tree topology for communications. Results show how the scalability of DLML was improved, using two clusters of 40 and 1024 processing units, and executing dynamic and static applications.

show abstract

An Efficient Dynamic Load Balancing Scheme for Distributed Simulations on a Grid Infrastructure

Cited by 29 publications

References 11 publications

Dynamic load redistribution based on migration latency analysis for distributed virtual simulations

Dynamic load redistribution based on migration latency analysis for distributed virtual simulations

Dynamic Load Balancing Using Grid Services for HLA-Based Simulations on Large-Scale Distributed Systems

Load Balancing Algorithms with Partial Information Management for the DLML Library

Contact Info

Product

Resources

About