Dear All,
I am doing some performance testing using NCBI BLAST and I am looking for a protocol and/or dataset that can be used as a reference for benchmarking. I already ran some tests but I am not sure I was doing the right thing in the right way. Had a look at the ancient paper by Shpaer (
http://dx.doi.org/10.1006/geno.1996.0614 ) but neither the aim of the article nor the data is really relevant from my point of view.
What I want:
- is to find bottlenecks and quasi-optimal parameter sets for different scenarios where BLAST (more precisely the BLAST sw suite: [t]blast[pnx], megablast, etc) is used.
What I did:
- downloaded the NCBI C++ Toolkit and compiled it for both production and profiling configuration (using GCC)
- downloaded the STS database set for blastn and put it into the RAM of my PC trying to minimize the IO overhead (apparently the sw is still IO bound but much faster than reading data from disk)
- ran a simple blastn search for 50, 500, 5K, 50K and 500K long nucleotide sequences - they were actually snippets from DS231696 (Fusarium data) to collect timing information.
What I got:
- profiling information for the default parameter set: the preliminar results are making sense, but I am still not sure I am on the right path.
I am doing some performance testing using NCBI BLAST and I am looking for a protocol and/or dataset that can be used as a reference for benchmarking. I already ran some tests but I am not sure I was doing the right thing in the right way. Had a look at the ancient paper by Shpaer (
http://dx.doi.org/10.1006/geno.1996.0614 ) but neither the aim of the article nor the data is really relevant from my point of view.
What I want:
- is to find bottlenecks and quasi-optimal parameter sets for different scenarios where BLAST (more precisely the BLAST sw suite: [t]blast[pnx], megablast, etc) is used.
What I did:
- downloaded the NCBI C++ Toolkit and compiled it for both production and profiling configuration (using GCC)
- downloaded the STS database set for blastn and put it into the RAM of my PC trying to minimize the IO overhead (apparently the sw is still IO bound but much faster than reading data from disk)
- ran a simple blastn search for 50, 500, 5K, 50K and 500K long nucleotide sequences - they were actually snippets from DS231696 (Fusarium data) to collect timing information.
What I got:
- profiling information for the default parameter set: the preliminar results are making sense, but I am still not sure I am on the right path.
Comment