Computational Tools for Efficient Large-scale Discrete Event Models

Colloq: Speaker: 
Kalyan Perumalla
Colloq: Speaker Institution: 
Georgia Tech, Atlanta, GA
Colloq: Date and Time: 
Thu, 2005-05-05 10:00
Colloq: Location: 
ORNL 5100-Auditorium
Colloq: Host: 
Jeffrey S. Vetter
Colloq: Host Email:
Colloq: Abstract: 
Lately, there is renewed interest in computational tools and methods in software testbeds to model large, complex systems. For example, infrastructure simulations for security are employing large-scale software testbeds of vehicle transportation systems, large-scale data networks, and others. The testbeds are modeled at great detail and executed on powerful computational platforms. In another example, new discrete-event methods in software testbeds are enabling orders of magnitude efficiency improvements in modeling large-scale physics phenomena (e.g., plasma simulations). At the heart of these testbeds are synchronization and runtime engines capable of efficient parallel execution and integration. This presentation will examine some of these computational methods & tools developed to sustain the scale, complexity and variety of large-scale models. As driving applications, two domains will be considered, namely, simulation of large-scale TCP/IP networks, and global kinetic simulations of earth's magnetosphere. Experimental results from the tools executing on large clusters and supercomputing platforms will be presented. A rollback-based technique using a novel reverse computation approach will be described that is shown to carry well across multiple domains.
Colloq: Speaker Bio: 
Kalyan Perumalla is a Research Faculty Member in the Systems research group at the College of Computing, Georgia Institute of Technology since 2000. He received his Ph.D. in Computer Science from the Georgia Institute of Technology in 1999. His interests include experimental research in parallel/distributed systems, parallel discrete event simulation, large-scale network modeling, network security and parallel combinatorial optimization. His co-authored papers won the Best Paper awards at the Parallel and Distributed Simulation Workshops, in 1999 and 2002. He has built several research prototype distributed/parallel systems, which have been distributed to researchers worldwide. Kalyan is currently involved as Co-PI in multiple large-scale systems projects sponsored by NSF and DARPA.