SPSpfam 2.3.2 Overview
SPSpfam shows dramatic performance gains for average query lengths of 200
amino acids or less. For ten aa length queries the performance gain
was 57! At 200 aa length queries the speedup is still about one order
of magnitude greater than Hmmpfam. This is significant, as SPSpfam
gives you the same throughput on ten processing nodes as Hmmpfam gives
you on one hundred nodes! The hardware cost savings for equivalent
throughput can be enormous.
Also, your performance will exceed the above numbers for query sets
larger than 16 queries. Some hardware accelerators advertise a performance
gain of 180x over Hmmpfam 2.2g, (equivalent to ~140x for version 2.3.2).
For short sequences SPSpfam approaches one-third to one-fourth that speed
for a fraction of the cost.
Appendix A (Benchmark Conditions)
System: Dual processor 2.2 Ghz Pentium
IV xeon chip set.
Memory: 1 Gbyte DDR Ram
OS:
Red Hat Linux version 7
Both processors were used for Hmmpfam and SPSpfam.
The HMM library was Pfam_ls 8.0 (5049 models)
All runs on both codes were timed on subsequent
runs after one short initial run.
The Hmmpfam binary from the HMMER Intel Linux distribution was used
for the comparisons.
The Unix time command was used to time
all runs.
Appendix B (Independent Work)
See the poster by Martin Gollery, "A Comparison
of HMMPfam Implementations" for an independent comparison of SPSPfam 2.2g at www.unr.edu/BRIN/pfamposter.pdf
Page 3
|
