IIT computer science researchers won the Best Paper Award at the 2015 Association for Computing Machinery Special Interest Group on Modeling and Simulation (ACM SIGSIM) Conference on Principles of Advanced and Distributed Simulation (PADS). The conference was held in London on June 10-12.
The paper, titled “FatTreeSim: Modeling Large-Scale Fat-Tree Networks for HPC [High-Performance Computing] Systems and Data Centers Using Parallel and Discrete Event Simulation,” was authored by IIT computer science Ph.D. student Ning Liu, undergraduate student Adnan Haider (CS 2nd Year), Assistant Professor Dong (Kevin) Jin, and IIT Distinguished Professor of Computer Science and IEEE fellow Xian-He Sun.
In the era of big data, many scientific and commercial applications are limited by the growing demands of processing and storing large-scale datasets, and researchers are working to overcome such limitations. Liu and the other IIT researchers built a large-scale simulation system for fat-tree networks, a widely used network topology in large-scale parallel and distributed computing system, and then ran extensive experiments with it. Fat-tree networks face new challenges with the advent of the era of extreme-scale computing, when systems feature millions of physical cores and potential billion-way concurrency.
Liu said, “Our simulation system, named FatTreeSim, enables the parallel and distributed computing community to explore and evaluate large-scale fat-tree networks in a time- and budget-efficient manner. We can scale the FatTreeSim system to a total of 32,768 cores on Argonne National Laboratory’s IBM Blue Gene system, Mira, which is the fifth-fastest supercomputer in the world. The largest scale of fat-tree network we simulated is at the scale of half million servers. Given the proper resources, FatTreeSim can potentially simulate even larger-scale fat-tree networks very efficiently.”
“This research is of particular interest to the simulation community and will have a significant potential impact in modeling and designing large-scale network architectures for data centers and HPC systems,” added Jin.
“PADS is the premiere conference for academic work in discrete-event simulation and modeling,” said Sun. “We’re pleased at the recognition the PADS community has given to this work.”
Also at PADS, Liu won the Ph.D. Colloquium Award for his Ph.D. research. Having the two awards go to the same group is an unparalleled honor in PADS’ 30-year history, according to Sun.