🚀 go-pugleaf

In <1hsmkiINNf05@rave.larc.nasa.gov> s.j.scotti@larc.nasa.gov (Stephen J. Scotti) writes:


| >| When looking at gr_osview, I have a program that uses about 50% of the
| >| cpu (indigo R4000 elan running 4.05F). The CPU wait line in gr_osview
| >| is empty, so if the cpu isn't waiting, why doesn't my program get 100%
| >| of the CPU?
| >| 
| >| I guess the less naive way of putting this is: What factors aren't captured
| >| between the CPU Usage and CPU wait?
| >| 
| >| [The program in question reads in a 16k chunks of stuff off of disk, messes
| >| around with them and writes them out. All to local disk.]
| 
| >It's hard to say just where it is waiting.  Running osview may be
| >helpful, as would 'timex -s yourprog'.  My guess is that it is indeed
| >waiting for the disk part of the time.
| 
| I have been running a structural analysis finite element code on a 4d35 with 
| similar problems.  I was the only user on the machine but could get no more 
| than about 60% of the CPU.  I too attributed this to disk accesses but when 
| we upgraded our memory from 16mb to 48mb, I found that I could get about 95% 
| of the CPU!
| I wonder if Dave Olson could explain that?

More buffering of files, and therefore less time spent waiting
for the disk, is my *guess*.  It could also be that you were
doing a lot more paging with less memory.  Both timex -s and
osview would help confirm or deny these theories.  Without
measurements, who knows?
--
Let no one tell me that silence gives consent,  |   Dave Olson
because whoever is silent dissents.             |   Silicon Graphics, Inc.
    Maria Isabel Barreno                        |   olson@sgi.com