Seeking Super-Linearity Speedup in birgHPC Environment
Abstract
Abstract
The aim of this research is to show whether a super-linearity speedup can be or cannot be achieved within normal personal computer in university laboratory with an effort-less cluster, based on a pre-boot execution environment (PXE) operating system. Super-linearity is a case of speedup where the time execution ratio of sequential versus parallel is more than the number of computing node. We create the High Performance Computing cluster environment based on Bioinformatics Research Group High Performance Computing (birgHPC), a remake of PelicanHPC. The environment is limited to 32 regular personal computers as PXE computing nodes, Pentium4 3.2GHz CPU. The head-node is equipped in Pentium4 3.2 GHz CPU, 2.5 GB RAM, 80 GB hard-disk. Basic Local Search Tool (BLAST) and Message Passing Interface BLAST (mpiBLAST) problems are conducted, searching sequences in 8.9GB unformatted database, within several environment settings. The challenge is to beat 7+ hours of computing time in a single machine of BLAST problem. It is concluded that super-linearity is a rare case that can be shown in simple limited environment. Even the experiment environment is tweaked: downsize the computing node RAM and upsize the searching sequence such that the cache effect can be significant.
Keywords: Super-Linearity, High Performance Computing, Bioinformatics, mpiBLAST
References
References
Altschul, S et al. Basic Local Alignment Search Tool, Journal of Molecular Biology. 1997; 215:403-410.
Darling A.E, Carey L and Feng W.C, The Design, Implementation and Evaluation of mpiBLAST, 4th International Conference on Linux Cluster 2003.
Chew, T.H. et al. birgHPC: Creating Instant Computing Cluster for Bioinformatics and Molecular Dynamics. Oxford Journals: Bioinformatics. 2011; 27(9):1320-1321
