FY2010 - Oak Ridge National Laboratory

Director’s R&D Fund—
Ultrascale Computing and Data Science
Perumalla, K. S. 2010. µπ: A Scalable and Transparent System for Simulating MPI Programs. ICST
International Conference on Simulation Tools and Techniques. Torremolinos, Italy.
Perumalla, K. S., and C. Carothers. 2010. Compiler-based Automation Approaches to Reverse
Computation. Reverse Computation Workshop (in conjunction with IEEE/ACM/SCS PADS'10),
Atlanta, GA, USA, IEEE Computer Society.
Perumalla, K. S., and S. K. Seal (2010). Reversible Parallel Discrete Event Execution of Large-scale
Epidemic Outbreak Models. IEEE/ACM/SCS International Workshop on Principles of Advanced and
Distributed Simulation. Atlanta, GA, USA, IEEE Computer Society (Best Paper Finalist).
05550
Computational Biology Toolbox for Ultrascale Computing
Igor B. Jouline, Bhanu Rekepalli, Andrey A. Gorin, and Christian Halloy
Project Description
Insufficient capability to translate the exponentially growing genomic data into useful knowledge is the
single most pressing grand challenge in biology. The goal of this project is to dramatically improve
biological function prediction by building new and improved models for mining genomic data. This goal
will be achieved by using most sensitive data mining tools organized in a robust, massively parallel
computational infrastructure. We will port these tools to a Cray XT5 supercomputer and adopt their usage
for developing cloud computing, thus enabling mining not only the existing genomic data, but also the
future data sets that will be larger by orders of magnitude.
There are two types of the project deliverables: (i) a newly developed toolbox containing most useful
computational biology software implemented for Cray supercomputers and (ii) a set of new and improved
models for biological function prediction that will become available worldwide through major national
and international databases. By investing in this project, ORNL will seize the opportunity to become a
leader in ultrascale computational biology and will position our team strategically to successfully compete
for major funding from the National Institutes of Health (NIH) and DOE.
Mission Relevance
This project aims at establishing ORNL as a world leader in dynamic knowledge discovery based on
capabilities for handling diverse genomic data. It will also contribute to developing focused research
communities in biology, because the computational biology toolbox developed by the project will be used
by a large community of biomedical scientists. This project also addresses the major problems of the
DOE (bioenergy) and NIH (human health), because improved biological function prediction is urgently
needed to solve these problems.
Results and Accomplishments
Nearly all deliverables planned for the Year 1 have been met or exceeded.
BLAST. We optimized the BLAST code. First we installed the BLAST tool on the Kraken supercomputer
and profiled the code to understand the I/O. The database broadcasting to all the nodes in the job was
optimized and then the I/O was optimized in two phases. First, a buffer was created in which all the input
query sequences were stored and a dynamic load balancing algorithm was designed to distribute the work
optimally between all the cores of the node. Second, the outputs from each core were put into separate
90