Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


6778.pdf (1.02 MB)

ETD Abstract Container

Abstract Header

Event List Organization and Management on the Nodes of a Many-Core Beowulf Cluster

Dickman, Thomas J
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1378196499

Abstract Details

2013, MS, University of Cincinnati, Engineering and Applied Science: Computer Engineering.
Parallel Discrete Event Simulation (PDES) has been widely studied for many years. One of the principle mechanisms for the distributed synchronization of concurrent event execution in a PDES is an optimistic method called the Time Warp Mechanism. Researchers at the University of Cincinnati have been studying the Time Warp mechanism for many years and they have developed an implementation of the Time Warp mechanism for execution on Beowulf clusters that is called WARPED. In recent years, the WARPED kernel has been modified to add integrated thread support for execution on multi-core and many-core Beowulf clusters. While this threaded WARPED kernel provides effective execution support for Beowulf clusters containing small multi-core processors, contention for the shared input event queue (or pending event set) on a node has a negative impact on the overall performance of the simulation once the number of threads in the nodes grows above 6-8. This thesis explores the design of the software architecture for the pending event set in the WARPED kernel. In particular, we examine a solution where the pending event set on each node is partitioned into multiple input queues (called LTSF queues) with disjoint subsets of the event processing threads bound to each LTSF queue to help reduce contention. While this reduces contention for the shared pending event set to managable levels, it can potentially lead to an unbalanced advancement of simulation time among the subsets of event processing threads. Thus, this thesis also examines the implementation of methods to dynamically adjust the partitioning of events distributed to the various LTSF queues on a node. Several algorithms and metrics are used to decide when and where to adjust the workload with improved run time results occurring as a result.
Philip Wilsey, Ph.D. (Committee Chair)
Fred Annexstein, Ph.D. (Committee Member)
Fred Beyette, Ph.D. (Committee Member)
79 p.
Computer Engineering
Time Warp; pending event lists; multi-core; threads; Beowulf clusters; parallel computing

Recommended Citations

Citations

  • Dickman, T. J. (2013). Event List Organization and Management on the Nodes of a Many-Core Beowulf Cluster [Master's thesis, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1378196499

    APA Style (7th edition)

  • Dickman, Thomas. Event List Organization and Management on the Nodes of a Many-Core Beowulf Cluster. 2013. University of Cincinnati, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1378196499.

    MLA Style (8th edition)

  • Dickman, Thomas. "Event List Organization and Management on the Nodes of a Many-Core Beowulf Cluster." Master's thesis, University of Cincinnati, 2013. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1378196499

    Chicago Manual of Style (17th edition)

ucin1378196499
521
© 2013, some rights reserved.Creative Commons License Event List Organization and Management on the Nodes of a Many-Core Beowulf Cluster by Thomas J Dickman is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. Based on a work at etd.ohiolink.edu.
This open access ETD is published by University of Cincinnati and OhioLINK.