[Bioclusters] Stranger Time to serial BLAST

[Tim Cutts]

On 23 Mar 2007, at 6:35 pm, Daniel Xavier de Sousa wrote:

> The difference using more memory, must be better. For
> exempla, there is little difference using 1 GB and 512 MB for a  
> database of
> 1.8GB (NR). The little difference is more accentuated to WUblast  
> 1.4 than NCIBI
> BLAST.

1)  You'd see a *much* more significant difference as soon as the  
machine has enough memory to cache the entire database in RAM.

2)  Your test is also of BLASTP.  I find that having sufficient RAM  
makes a lot more difference to BLASTN than to BLASTP - I suppose this  
is isn't surprising since nucleotide sequences are generally longer,  
so both the database and the DP matrix will be larger.

>
> 2 – What can I do to get more difference when I change my
> RAM memory?

Make sure you have enough memory that the entire database is resident  
in RAM, and you'll see a dramatic change in performance, especially  
when performing multiple queries one after another of the same database.

We buy all our machines now with 2GB of RAM per core, which means on  
current dual-socket dual-core machines we always have at least 8GB  
RAM, and it makes an enormous difference (even the users noticed!)  
since it can cache an entire mammalian genome in RAM.

Tim