Talk by Jaliya

Date: October 14th 2008
Time: 1.00 p.m.
Room: 100B

Title: Performance of Runtime Environments for Data Intensive Supercomputing

Abstract: Computation and data intensive scientific data analyses are
increasingly prevalent. In the near future, it is expected that the data
volumes processed by applications will cross the peta-scale threshold, which
would in turn increase the computational requirements. Two exemplars in the
data-intensive domains include High Energy Physics (HEP) and Astronomy. Data
volume is not the only source of compute intensive operations. Clustering
algorithms used in many domains such as biology and chemistry are especially
compute intensive even though the datasets are comparably smaller than the
physics and astronomy domains. Efficient parallel/concurrent algorithms and
implementation techniques are the key to meeting the scalability and
performance requirements entailed in such scientific data analysis. Most of
these analyses can be thought of as a Single Program Multiple Data (SPMD)
algorithm or a collection thereof. These SPMDs can be implemented using
different parallelization techniques such as threads, message passing,
map-reduce, and workflow technologies. Additionally, the direct and virtual
hardware environments create another dimension for the overall performance
of these applications. The goal of this research is to evaluate the
performance of various runtimes using real scientific applications on direct
and virtual hardware environments and understand how the features such as
scalability, fault-tolerance, and dynamic flexibility, provided by these
runtimes and the hardware environments can be used to improve the overall
performance of the scientific applications. Finally, I will use this
knowledge to derive a set of architectural recommendations for these
runtimes.

Thanks,
Jaliya

--
Jaliya Ekanayake
http://www.cs.indiana.edu/~jekanaya/
Phones Home:812-339-5418 Mobile: 812-606-0561 Lab:812-856-0758