In article <9405171436.AA21861 at ugene1.abbott.com>,
Tim Bolling <bollingt at ugene1.abbott.com> wrote:
>>>BENCHMARKS
>>Some recent Alpha AXP performance benchmarks that may be of interest are:
>>>FASTA Benchmark (run with cooperation of MIT/Whitehead)
>>_______________________________________________________
>>SPARC10/40 1.20 sec
>>AlphaAXP OSF/1 3000-600 .37 sec
>>AlphaAXP OSF/1 3000-800 .31 sec
>>Not that I'm refuting this data but I have to say that the actual
>times seem a bit suspicious to me. At our site, the most common use
>of FASTA is to query a sequence against one of the public databanks.
>The time necessary to do this search on a 690MP is more on the order
>of 20 min not 1.20 sec. Much of that time is spent reading data from
>the disk so the type of disk, its controller, and the amount of
>memory, etc., in addition to the CPU factor into the real-time that
>elapses when one executes a search program. Any search done in 1.2
>sec. is obviously not searching one of the public databanks
>(ie. GenBank, EMBL, etc) and thus, is not representative a typical
>"real-life" use of FASTA.
(These are typical SwissProt protein database searches I suspect>0
>On the otherhand, the BLAST times (not included in this message) look pretty
>representative of actual times I've experienced. If the FASTA results were
>to hold in a more extensive search, then it would appear that the Alpha
>might indeed be 2-4 times faster than the SPARC10.
It is important to distiguish between the FASTA that GCG
distributes, which I call SLOWA, and the "native" FASTA that I
distribute. The GCG library reading functions are very slow; the
native version uses less than 10% of its time reading libraries. In
my tests, FASTA with ktup=2 is about the same speed as BLASTP on the
same dataset (although BLASTP is more sensitive and more selective).
These times are what I would expect base on my experience with
my AlphaAXP 3000-300.
Bill Pearson
