[Bioclusters] Xserve G5 memory

David Huen bioclusters@bioinformatics.org
Tue, 5 Oct 2004 20:43:24 +0100


On Tuesday 05 Oct 2004 19:33, Ruotti, Victor wrote:
> Hi,
> I agree with Chris on this. BLAST databases are flat files indexed by the
> formatdb program. Does anybody know whether the latest version of
> formatdb allows you to index BLAST databases directly into memory? I
> don't remember whether formatdb does this or not. Can somebody comment on
> other methods to put BLAST databases into memory?
>
I am unclear of the advantage of formatdb directly to memory - it will not 
persist anyway.

I suspect the effect observed is from the propensity or otherwise of the OS 
to cache recently-read files blocks in memory. With Linux, this can often 
be observed just by typing "cat foo.file > /dev/null" and using the file 
immediately thereafter.  Also with multiple CPUs (in the 8-way IBM) and 
presumably blast jobs running on many of them almost continuously, the 
cached blocks remain in the cache.  Perhaps OS-X file cache purges/does not 
cache recently read blocks more readily.  I assume you do have plenty of 
RAM so cached disk blocks don't get purged.  Those BLASTDB files can get 
quite large and will be at least partly purged if memory pressure occurs.

Just my $0.02 worth.

Regards,
David Huen