[ssml] problems with blastpgp on x86_64 machines

[Mensur Dlakic]

Hi Kevin,

I don't have a direct answer to your problem, yet I have seen the warning 
below (and few others). As far as I can tell, they are exactly that, just 
warnings, so your problem should be elsewhere.

In a separate private message, I'll attach a little Linux program called 
readchk2. It reads PSI-BLAST checkpoint files and outputs a text file that 
contains the sequence from which the checkpoint was generated and the 
amino-acid frequencies stored in .chk files. If something is wrong with the 
checkpoint files, which you seem to be using from the example below, this 
output may be helpful. For example, I have seen checkpoint files that have 
the sum in some rows quite a bit different from 1, say 0.9, and that tends 
to break PSI-BLAST. In my experience, it is normal if the sums differ from 
1 by 0.005 or so because of rounding errors.

Though I suspect that you have this possibility covered, the problem could 
be in big- vs small-endian checkpoints. I'll also include a program 
swapchk2 which swaps the bytes in .chk files between the two architectures, 
so if you run it twice on the same file the values should be back to the 
original. Beware, the program overwrites the file in question, so make sure 
you have a copy.

Finally, if anyone else is interested in these programs, I'd be glad to 
share them.

Cheers,

Mensur


At 02:39 PM 8/23/2007, Kevin Karplus wrote:

>I've been having trouble with blastpgp (the program that is the core
>of psi-blast) crashing on my cluster of x86_64 machines.  The crash is
>not consistent---rerunning the same jobs can cause everything to
>complete fine, but is frequent enough to be a serious problem (2/3 of
>the jobs in a recent run crashed---on rerunning about 1/3 of the
>previously crashed ones completed ok).
>
>About the only symptom I'm seeing besides the crashes is
>[blastpgp] WARNING:  [000.000]  Failed to access string index ISAM Error 
>code is -5
>but that isn't even coming from the times that blastpgp is crashing!
>
>The same jobs run on older 32-bit machines do not seem to crash.
>
>I'm running fairly recent (Mar 2007?) of the software.
>
>Anyone have any suggestions?

==========================================================================
| Mensur Dlakic, PhD                | Tel: (406) 994-6576                |
| Department of Microbiology        | Fax: (406) 994-4926                |
| Montana State University          |                                    |
| 109 Lewis Hall, P.O. Box 173520   | http://myprofile.cos.com/mensur    |
| Bozeman, MT 59717-3520            | E-mail: mdlakic at montana.edu        |
==========================================================================