New Parallel Scheme of HMMer PFAM

1
New Parallel Scheme of HMMer PFAM

• Weirong Zhu
• Niu Yanwei

2
(No Transcript)
3
(No Transcript)
4
(No Transcript)
5
Advantages of our new parallel scheme further
possible improvement

• Suppose we have n profiles in DB, k
  sequences in seq file, each time t sequences are
  transferred from node to node. Each profile is X
  bytes, each sequence is Y bytes

Further improvement Paralleling post computation
on master node after collecting result from
slave nodes.
6
Some Simple Timing Result

• I embeded my timing code into original hmmpfam
  sequential code (no pthreads, no pvm)to see the
  cost of read data from HMM profile DB,and SEQ
  file.
• --------------------------------------------------
  ---------------
• hh.db 112.833 K, hh1.seq 57.123 K
• Total execution Time 5.889 s
• Total Time for read HMM DB 2.199 s (37.3)
• Total Time for read SEQ file 0.008 s
• --------------------------------------------------
  -----------------
• hh.db 112.833K, hh2.seq 282.747K
• Total execution Time30.814 s
• Total Time for read HMM DB12.305 s (39.9)
• Total Time for read SEQ file 0.016 s
• --------------------------------------------------
  -----------------
• hh.db 112.833K, seq.cgi 10.524569M
• Total execution Time966.029 s
• Total Time for read HMM DB330.375 s (34.2)
• Total Time for read SEQ file 0.660 s