In article <4vpko6$9ii at lyra.csx.cam.ac.uk> tjrc1 at mole.bio.cam.ac.uk (Tim Cutts) writes:
In article <kjenkwd1mb.fsf at alpha0.bioch.Virginia.EDU>,
William R. Pearson <wrp at alpha0.bioch.virginia.edu> wrote:
>This version correctly threads on multiprocessors Solaris2.5.1 (a
>change between Solaris2.4 and 2.5.1 apparently caused the program not
>use more than 50% of the processors - this has been fixed).
Isn't that a good thing though? On a multi-user machine, you don't
want a single process to just grab every single processor without any
regard to interactive performance. We have had several problems on
our site with FASTA destroying interactive performance on our main
machine, which also has to function as a mail host for 1500 people.
(Is this my FASTA or GCG's FASTA? GCG's is considerably slower and
more resource intensive.)
It seems that under IRIX 5 at least, fasta has a tendency to grab 80%
CPU even when niced as much as possible, which is just terrible.
Allocating CPU time slices appropriately is a job for the OS
scheduler, not for FASTA. If someone wants to run a dedicated FASTA
machine, FASTA should use all the CPU's. To run FASTA on a subset of
CPU's, fasta3_t has a compile time option that allows you to set the
maximum number of threads that will be used).
If FASTA is using 80% of a machine when nice'd one would hope that
there was not much else that the CPU was waiting to do. Or "nice" is
not working properly.