<br><br><div class="gmail_quote">---------- Forwarded message ----------<br>From: <b class="gmail_sendername">Yaacov Yesha</b> <span dir="ltr"><<a href="mailto:yayesha1@gmail.com">yayesha1@gmail.com</a>></span><br>Date: Thu, Sep 20, 2012 at 3:00 AM<br>
Subject: CSEE Colloquium on September 21 at 1:00 pm - Simon on task scheduling for high performance distributed systems<br>To: <a href="mailto:csee-colloquium-out@cs.umbc.edu">csee-colloquium-out@cs.umbc.edu</a><br><br><br>
<br><br><div class="gmail_quote">---------- Forwarded message ----------<br>From: <b class="gmail_sendername">Tim Finin</b> <span dir="ltr"><<a href="mailto:finin@cs.umbc.edu" target="_blank">finin@cs.umbc.edu</a>></span><br>
Date: Wed, Sep 12, 2012 at 8:24 AM<br>
Subject: [Csee-faculty-tt] [CSEE-colloq] talk: Simon on Task Scheduling for High Performance Distributed Systems, 1pm Fri 9/21<br>To: <a href="mailto:csee-colloquium-out@cs.umbc.edu" target="_blank">csee-colloquium-out@cs.umbc.edu</a><br>
<br><br> CSEE Colloquium<br>
<br>
A Novel Dynamic Task Scheduling Environment<br>
for High Performance Distributed Systems<br>
<br>
Tyler Simon<br>
Faculty Research Assistant, UMBC<br>
<br>
1:00pm Friday, 21 September 2012, ITE 227, UMBC<br>
<br>
The number of concurrently executing tasks required for a single<br>
application to perform at the petascale is on the order of hundreds of<br>
thousands. Given current manycore hardware trends, future peta- and<br>
exa-scale class systems will require applications to run tasks on the<br>
order of hundreds of millions to billions. To address the problem of<br>
creating, running and managing jobs of this scale, both from a system<br>
user and administration perspective we have developed, ARRIA, an<br>
Autonomic Runtime for Resource Intensive Applications. ARRIA uses a<br>
decentralized bag of tasks and workload scheduler that increases<br>
individual job priorities based on weighed factors that are of<br>
interest to the application programmer or the system administrator.<br>
ARRIA is designed to run millions of independent tasks reliably and<br>
efficiently without explicit message passing from the user. In<br>
previous work, using the ARRIA scheduler for scientific MapReduce<br>
workloads, we have shown a 2.1x speedup over the Hadoop Fair Share<br>
scheduler. We investigate novel scheduling parameters and strategies<br>
that guarantee efficient job execution for a wide range of realistic<br>
and simulated workloads with both user and administrator objectives,<br>
such as increased throughput and maximized utilization with minimal<br>
wait times for specific job classes. Finally our experiments<br>
investigate the long tail phenomenon for mixed workloads and the<br>
overheads incurred for increased system size.<br>
<br>
Mr. Simon has undergraduate degrees in Computer Science and Philosophy<br>
with a Master of Science in Computer Science from the University of<br>
Mississippi, he is currently pursuing a PhD in Computer Science at the<br>
University of Maryland Baltimore County. Mr. Simon has worked<br>
professionally in the high performance computing (HPC) field for over<br>
a decade. In 2005 he earned a Department of Energy graduate research<br>
fellowship at Oak Ridge National Laboratory, where he worked for in<br>
the Computer Science and Mathematics Division developing and<br>
implementing the Freeloader distributed storage system. Mr. Simon has<br>
worked as a computational scientist for the Department of Defense High<br>
Performance Computing Modernization Office based at the U.S. Army<br>
Engineer Research and Development Center in Vicksburg, MS, evaluating<br>
both current and future HPC system requirements for applications of<br>
interest to the Department of Defense. Since 2009 Mr. Simon has been<br>
a computational scientist and manager of HPC user services at the NASA<br>
Center for Climate Simulation at Goddard Space Flight Center and is<br>
currently a Faculty Research Assistant at the University of Maryland<br>
Baltimore County working at the NSF Center for Hybrid Multicore<br>
Productivity Research. Mr. Simon’s research involves the study of<br>
dynamic distributed runtime environments, parallelization strategies<br>
and scheduling of large scale scientific applications for current<br>
petascale and future HPC architectures.<br>
<br>
-- For more information and directions see <a href="http://bit.ly/UMBCtalks" target="_blank">http://bit.ly/UMBCtalks</a> --<br>
--<br>
For info and options, see <a href="http://lists.cs.umbc.edu/mailman/listinfo/csee-colloquium-out" target="_blank">http://lists.cs.umbc.edu/<u></u>mailman/listinfo/csee-<u></u>colloquium-out</a><br>
______________________________<u></u>_________________<br>
Csee-faculty-tt mailing list<br>
<a href="mailto:Csee-faculty-tt@cs.umbc.edu" target="_blank">Csee-faculty-tt@cs.umbc.edu</a><br>
<a href="http://lists.cs.umbc.edu/mailman/listinfo/csee-faculty-tt" target="_blank">http://lists.cs.umbc.edu/<u></u>mailman/listinfo/csee-faculty-<u></u>tt</a><br>
</div><br>
</div><br>