Parallel BLAST multithreading, old and new question

[Bachbioinfo]

Hello,

I am wondered what is the best way to run blastn on a huge query file (5G ) against small database (SILVA or RDP) in parallel CPUs ?

1- split queries into parts which each part is smaller then database size, then run the blastn on all parts at the same time. Then use mpiru or qsub -pe of (#$ -pe smp 8G ; #$ -l h_vmem= ? ! 8x8) on each part. I ignore until now how to run the blastn on all parts at the same time, not just doing a "while" or "for" into the parts list

2- Use it simply with -num_threads option of blastn on (each part of course)
3- use "-task megablast" in blastn

I have the same question when we have a huge database like nr or nt.
Please share with me your views, help, remarks into the aim of improving these questions and finding right answers

thanks

[mikesh]

As far as I remember, -num_threads option is actually not working as expected, and blast runs in a single thread most of time. What I usually do (for the same type of problem): split query to #threads parts and run parallel on them. Sometimes it is wise to do redundancy elimination, something like CD-HIT, or at least combine identical reads.

[maubp]

If your query file has 10,000s of sequences, break it up into sub-files, e.g. files of 1000 sequences each. Then run a separate blastn for each query file, and if need be combine the output.

[Bachbioinfo]

I see, I think spiting the query into multi-files is the best solution.
the option -num_threads i not working correctly

Thanks

[maubp]

I think the -num_threads option works better on large databases.