Georgia Tech Student Research Increases Human Genome Indexing Speed by 110x and Advances Internal Memory Capacity

*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************

Discoveries are two of four finalists for “Best Student Research Paper” at Supercomputing ’15 conference

Contact

Tara La Bouff, 404.769.5408

 

Sidebar Content
No sidebar content submitted.
Summaries

Summary Sentence:

Four doctoral students make advances in high-performance computing (HPC) that outperform other common approaches in HPC today.

Full Summary:

No summary paragraph submitted.

AUSTIN, Texas Monday, Nov. 16, 2015 — Four doctoral students comprising two research projects in the School of Computational Science & Engineering at the Georgia Institute of Technology are finalists for “Best Student Research Paper” at Supercomputing ’15, the International Conference for High Performance Computing, Networking, Storage and Analysis.

For the first project, the finalists developed a fast algorithm; in the second, they created a GPU-based framework that can process large graphs exceeding a device’s internal memory capacity. Each demonstrably outperforms other common approaches in high-performance computing today. A winner will be announced at the conference in Austin, Texas.

“Both of these are outstanding projects and evidence of the future leaders who are already keeping pace with and solving the most challenging problems in science, engineering, health and the social domain,” says David A. Bader, professor and chair of the School of Computational Science & Engineering.

New Human Genome Indexing Algorithm for Parallel Distributed Memory

In his work, “Parallel Distributed Memory Construction of Suffix and Longest Common Prefix Arrays,” PhD Candidate Patrick Flick, working with Professor Srinivas Aluru, created parallel algorithms for distributed-memory construction that are 110x times faster than the best method running on a sequential, single computer. Flick, using the human genome as a racetrack to test his speed, indexed it in only 7.3 seconds using his distributed-memory algorithm running on 1024 Intel Xeon cores.

“Bioinformatics is an example of a scientific field that is extremely data intensive; speed matters and speed helps,” he says. “We are not aware of any other parallel suffix array or suffix tree construction algorithms which achieve speedups this high.”

It is believed to be the first algorithm and implementation that uses this approach for distributed-memory parallel systems.

“At this stage the code is offered as an open-source library that can be used within parallel applications,” Fick adds. “We hope that it finds adaptation within bioinformatics research. We are now working on a user-friendly interface that can be used by bioinformaticians to replace older (and slower) tools.”

Next, Flick is working on a journal paper that includes some more improvements and additional techniques, and further showcases their algorithms on real applications. Patrick also authored another paper at Supercomputing 2015 with fellow students Chirag Jain and Tony Pan about how to partition large graphs that arise in metagenomics, another data-intensive application area.

Processing Large-Scale Graphs

In their paper titled “GraphReduce: Processing Large-Scale Graphs on Accelerator-Based Systems,” PhD Candidates Dipanjan Sengupta, and Kapil Agarwal developed a scalable framework (dubbed “GraphReduce”) to process large graphs that exceed a device’s GPU memory.

“GraphReduce can accelerate the analysis of graphs with billions of edges, operating at speeds much faster than similar operations on CPUs, and programmed in ways that are accessible to those who are not typically experts in GPU programming,” Sengupta says.

It provides a logic for processing “shard stores” based on choice of interval, number and sizes of shards, and how to order the edges in each shard. A “graph layout engine” then defines the layout of the data by sorting in-edges by their destination and out-edges by their source.

“One of the interesting results is that saturating the available bandwidth and overlapping data transfer with computation was able to hide a large amount of overhead, resulting in huge performance benefits,” Agarwal adds.

Most methods process dynamic graphs (like those of a social network which are continually changing over time) by storing static versions of the graph and then repeatedly running analysis on them. To address this, Sengupta next is working on an open-source framework with a Fortune 500 company to boost processing.

About the Georgia Tech College of Computing

The Georgia Tech College of Computing is a national leader in the creation of real-world computing breakthroughs that drive social and scientific progress. With its graduate program ranked 9th nationally by U.S. News and World Report, the College’s unconventional approach to education is expanding the horizons of traditional computer science students through interdisciplinary collaboration and a focus on human-centered solutions. For more information about the Georgia Tech College of Computing, its academic divisions and research centers, please visit http://www.cc.gatech.edu

 

Additional Information

Groups

College of Computing

Categories
Research
Related Core Research Areas
Bioengineering and Bioscience, Data Engineering and Science
Newsroom Topics
No newsroom topics were selected.
Keywords
Genomics, high-performance computing, supercomputers
Status
  • Created By: Tara La Bouff
  • Workflow Status: Published
  • Created On: Nov 16, 2015 - 11:41am
  • Last Updated: Oct 7, 2016 - 11:20pm