[ssml] BLAST fastacmd failed to fetch sequence from database stored on PVFS2 filesystem

[Yun He]

Yes, to decrease the network traffic and the server IO, I had copied the 
databases to every node's local disk. Here I just want to test the 
performance of PVFS2 since it is said it shows excellent performance in 
large-IO environment. Yestoday I used FORMATDB to format the nr database in 
local disk and in PVFS2. The procedure was successful in local disk, but FAIL 
in PVFS2.

This is the log file of FORMATDB in local disk with successful messages:
========================[ Dec 6, 2007  9:07 PM ]========================
Version 2.2.17 [Aug-26-2007]
Started database file "nr"
Closing volume nr with 2976302 sequences, 999,999,232 letters(.psq file = 
1002976321 bytes; .phr file = 846550430 bytes)
Formatted 2976302 sequences in volume 0
Version 2.2.17 [Aug-26-2007]
Started database file "nr"
Formatted 2702180 sequences in volume 1
SUCCESS: formatted database nr


This is the log file of FORMATDB in PVFS2 complaining the errors:
========================[ Dec 6, 2007  9:28 PM ]========================
Version 2.2.17 [Aug-26-2007]
Started database file "nr"
Closing volume nr with 2976302 sequences, 999,999,232 letters(.psq file = 
1002976321 bytes; .phr file = 846550430 bytes)
ERROR: [000.000] Failed to create index: ISAMErrorCode -5.

Removed single-volume database nr
FATAL ERROR: [001.000] Fatal error when adding sequence to BLAST database.


On Thursday 06 December 2007 01:05, Kevin Karplus wrote:
> For speed reasons, we copy the databases to the individual nodes of
> our cluster when we do the weekly download.  Even a small cluster can
> bring a file server to its knees if all the nodes are running blast.
> have you check the logs on the file server to make sure that it is not
> getting overwhelmed---many file servers start behaving very badly
> indeed when they are overloaded. (not just slow, but dropping requests.)

-- 
Yun He   Ph.D.
National Laboratory of Biomacromolecules
Institute of Biophysics, Chinese Academy of Sciences
15 Datun Road, Chaoyang District
Beijing 100101
China
Tel: +86 010 6488 8487
E-mail: jarod at nlbmol.ibp.ac.cn
  or    jarod at spg.biosci.tsinghua.edu.cn