Login
Talks & Seminars
Title: Data movement optimizations in the pre-exascale era
Dr. Preeti Malakar , Argonne National Laboratory, USA
Date & Time: September 1, 2016 14:30
Venue: Conference Room, 01st Floor, C Block, Dept. of CSE, Kanwal Rekhi Bldg.
Abstract:
Computational science helps in solving fundamental problems in various scientific fields such as climate and weather, material science, drug discovery and many others. The complexity of the mathematics and the large number of calculations involved in modeling these real world scientific problems necessitates the use of high-performance computing. This has led to the development of current petascale supercomputers which have enabled the execution of transformative scientific simulations at unprecedented scales. An important step in the life-cycle of scientific discovery is data analysis and visualization of the simulation output that enables the scientists to derive insight from their simulations. While the computational power of supercomputers continues to increase with every generation, network interconnects and I/O systems have not kept pace, resulting in significant performance bottlenecks for data movement in these systems. Therefore, traditional post hoc analysis of huge simulation output will soon become inconceivable for the anticipated exascale applications. Thereby, concurrent data analysis is desirable and necessary. In this talk, I will describe some of my recent efforts towards optimizing concurrent data analysis workflows for large scale simulations, taking into account the various resource constraints such as processing speed, storage bandwidth, and network bandwidth. Next, I will describe my work on topology-aware data movement at scale. Researchers can achieve lower solve time for their science by reducing the data movement time. Optimizing data movement in exascale systems is challenging due to complex network topologies, proprietary routing algorithms, and dynamic application communication patterns. In this talk, I will outline strategies to reduce the parallel read/write times in large-scale scientific applications. I will illustrate a route-aware and load-aware algorithm to modify the existing I/O node assignment in the Blue Gene/Q supercomputer. This mitigates network contention and reduces write times by an average of 60% over the default MPI independent I/O on up to 8192 Blue Gene/Q nodes. The algorithm routes 1.4x fewer messages through the I/O nodes on the Blue Gene/Q.
Speaker Profile:
Preeti Malakar is a Postdoctoral Appointee in the Argonne National Laboratory. Her current work involves modeling simulation-time analysis for large-scale simulations and topology-aware data movement for applications running on some of the fastest supercomputers. She received her PhD from the Indian Institute of Science, Bangalore in 2014. As part of her thesis, she worked on improving throughput of weather simulations with multiple regions of interest, which was one of the finalists for the Best Student Paper in ACM/IEEE Supercomputing 2012. Preeti received the Best Paper Award in the Student Research Symposium at IEEE International Conference on High Performance Computing, 2009. She was awarded the Google India Women in Engineering Award (now Google Anita Borg Scholarship) in 2011 and the TCS Research Fellowship during her PhD.
List of Talks

Webmail

Username:
Password:
Faculty CSE IT
Forgot Password
    [+] Sitemap     Feedback