Skip to Main Content U.S. Department of Energy
PNNL: High Performance Computing



We evaluated the feasibility of designing a scalable parallel framework that can achieve orders of magnitude higher pairwise sequence alignments per second (PSAPS) performance than contemporary software. As a result, we developed a TASCEL-based parallel approach to perform large-scale homology detection using work stealing. i) We chose the all-against-all model not only for its broad scope of applications, but also because it occupies an upstream phase in most sequence analysis workflows; ii) To ensure high quality of the output, each pairwise sequence alignment (PSA) is evaluated using the optimality-guaranteeing Smith-Waterman algorithm; iii) We used protein/putative open reading frame inputs from real world data sets to capture a more challenging use-case where a skewed distribution in sequence lengths can cause nonuniformity in PSA tasks; and iv) To the best of our knowledge, this effort represents the first use of work-stealing for this problem domain. We demonstrated (ParGraph'12) homology detection at the largest scale in number of cores (x 10^5 cores), and we report the highest PSAPS performance (2.42x10^7 at 120,000 cores) reported for optimal homology detection --- roughly two orders of magnitudes higher than the top PSAPS reported previously (6.59x10^5 on 2048 cores).

Daily JA, S Krishnamoorthy, and A Kalyanaraman. 2012. "Towards Scalable Optimal Sequence Homology Detection." In 19th International Conference on High Performance Computing (HiPC), December 18-22, 2012, Pune, India. Institute of Electrical and Electronics Engineers, Piscataway, NJ. doi:10.1109/HiPC.2012.6507523

Daily JA. 2015. "Scalable Parallel Methods for Analyzing Metagenomics Data at Extreme Scale." PNNL-24266, Pacific Northwest National Laboratory, Richland, WA.

Contact: Jeff Daily

Optimizing legacy HPC applications

Preparing codes for next generation supercomputers systems is anticipated to require significant changes to the optimization strategies employed in stable HPC applications. This work employed term-rewriting transformations to optimize such applications. Among other optimizations, the application-specific term-rewriting transformations designed in this work transformed the benchmarks considered to use TASCEL for dynamic load balancing.

A. Panyala, D. Chavarria, and S. Krishnamoorthy. "On the use of term rewriting for performance optimization of legacy HPC applications". ICPP'12

Contact: Daniel Chavarria

Site Links

Our Research

Past Research