Original Question:
I have an AlphaServer 1000A running Digital Unix 4.0B. A user is running a
process, paup, which takes up 90% or more of the CPU. This process runs
for days! Is this OK for one process to take up all the CPU? I've tried to
renice the process, but that does not seem to make any difference.
Thanks for all the speedy responses!:
Dana McClure
Chad Price
Peter Stern
Girish Phadke
Jean-Loup Risler
Jim Belonis
Richard Bemrose
Dr. Tom Blinn
Michael Matthews
Ram Rao
Jeffrey Hundstad
Rob McCauley
Alan
SUMMARY:
First thing to do is to contact the user and verify that it is a legitimate
process and not a runaway process (which I had done).
Basically, the process is using all the CPU because it can and that is OK
because there is little demand of the CPU from the other processes. I have
noticed that it does give up some of the CPU when another process needs it.
Renice sets the priority on the process lower compared to other things.
Teach the user how to renice the process so other process with higher
priority will run first.
Other comments of note:
Ram Rao: V4.0D has a class scheduler which allows restricting the
percentage of the CPU that can be consumed by a process.
Richard Bemrose: Perhaps you should consider installing and running a
Network Queueing System (NQS) like GNQS;
http://www.shef.ac.uk/~nqs
-or-
http://www.gnqs.org (available soon)
Jean-Loup Risler: I am not surprised that PAUP eats your CPU for days. Other
phylogeny programs would do the same as well. My last run took
more than 200 CPU hours ....
There is a nice (!) program called "autonice" which
allows you to automagically define a nice value for any process, based
on the user name and/or the name of the process. Hence with autonice
you can systematically force PAUP to run with a high nice value.
The program is available on
ftp://ftp.ba.cnr.it/pub/users/massimo/autonice-0.5
For those interested in the original version, see:
ftp://ftp.mrc-bbc.ox.ac.uk/pub/dec_alpha/autonice-x.xx.tar.gz
Chad Price:
You are obviously running a system which supports molecular biologists and
molecular systematicists. I assume you also have the GCG/Wisconsin Pagage
loaded? And if not, probably should (it costs about $3K per year and
another $2.4K/year for the database updates).
PAUP is attempting to solve an NP complete problem. If the problem is
chosen carefully, the heuristics used will in fact allow it to complete.
If the user has not chosen carefully, then the program may run for weeks or
months before either you reboot, or one of you kills the job. As an
experiment, I started a paup job with 14 sequences and told it to do an
exaustive tree search and 100 iterations of the bootstrap. After 2,000
hours on a Sun hyperSparc system, I gave up and killed it, as I will be
rebooting in the not too distant future.
The way to get it to take less time (for a given number of sequences to
find a tree for) is to suggest to the user that they do NOT do an exaustive
search. An exausitive search does not really appreciably increase the odds
of finding an optimal solution, but does cause the program to run in
exponential time based on the number of (DNA or protein) sequences and the
number of differences between them. On the other hand, they may not be
doing an exaustive search, but may simply be anaysing a lot of sequences
and/or a set of very long or quite different sequences. If this is the
case, it's going to take a long time no matter what.
Email me if you want more details of what is going on there (or call). I've
been supporting Sequence Analysis software for about 5 years now.
Ellen Davis
Ellen.Davis_at_uc.edu
Received on Fri Apr 24 1998 - 16:11:29 NZST