Re: RHEL high percentage of processor time from system?

From: Ben (jbenjam_at_gmail.com)
Date: 10/16/05 

Date: 15 Oct 2005 19:56:49 -0700

The disk i/o is actually pretty low ... I really don't think it should
be an issue. No other processes are using CPU time (except 0.1% by
kjournald). Could be the filesystem...
disks aren't close to full yet, but they are regular IDE as far as I
can tell. (wouldn't this show up under "i/o wait" and the other values
from "top"?)

The Xeons are 2.4 GHz each if I recall correctly, and the FC3 box is 1
Intel P4 3.4 GHz HT.

Unfortunately I cannot stop these processes now, it will be two days or
so before they are done. I'll try some very simple tests, e.g.,
fortran vs. c vs. matlab with no disk i/o, fortran vs. c vs. matlab
with heavy disk i/o. I'm including Matlab in there to see how it
compares, and I have programs that do the same thing but are written in
all three "languages". I'll post back when I find out more
information! Thanks for the ideas.

On the FC3 box:
>vmstat 1 10
procs -----------memory---------- ---swap-- -----io---- --system--
----cpu----
 r b swpd free buff cache si so bi bo in cs us
sy id wa
 4 0 944092 58512 73868 97752 0 1 12 12 9 7 33
9 58 0
 4 0 944092 63056 73868 92552 0 0 0 72 1066 861 78
21 1 0
 4 0 944092 60688 73868 95672 0 0 0 0 1043 677 80
19 1 0
 4 0 944092 57032 73868 98792 0 0 0 0 1059 602 78
20 1 0
 4 1 944092 63944 73868 91772 0 0 0 13312 1117 529 69
30 0 0
 4 0 944092 60360 73876 95404 0 0 0 804 1171 524 77
22 1 0
 4 0 944092 75080 73876 80584 0 0 0 0 1042 689 78
21 1 0
 4 0 944092 69256 73876 86044 0 0 0 0 1062 617 69
30 1 0
 4 0 944092 64128 73876 91504 0 0 0 0 1043 697 74
25 1 0
 4 1 944092 64072 73876 91244 0 0 0 16036 1169 714 74
25 0 0

On the RHEL box:
>vmstat 1 10
procs memory swap io system
   cpu
 r b swpd free buff cache si so bi bo in cs us
sy id wa
 0 2 0 109120 418492 756524 0 0 0 5 8 7 10
8 9 1
 4 0 0 113040 418492 757760 0 0 0 360 207 1078 23
54 10 13
 4 0 0 110900 418492 761752 0 0 0 0 105 51 32
68 0 0
 4 0 0 107180 418492 765732 0 0 0 0 105 49 34
66 0 0
 4 0 0 103192 418492 769720 0 0 0 0 107 51 36
64 0 0
 4 1 0 91508 418492 773312 0 0 0 18468 304 62 50
46 0 4
 4 0 0 96884 418492 772596 0 0 0 516 285 2244 33
49 7 10
 4 0 0 93116 418492 776624 0 0 0 0 105 51 32
68 0 0
 4 0 0 89232 418492 780516 0 0 0 32 116 73 51
49 0 0
 4 0 0 85468 418492 784284 0 0 0 16 109 62 59
41 0 0

Relevant Pages

  • Re: How to change rsync to run upon rsh ?
    ... and therefore outweigh any kind of disk I/O. ... I can measure how much time (CPU time and real time) is used for ...
    (comp.security.ssh)
  • Re: Why no single dual-processor Mac Pro?
    ... Disk i/o, but the CPU time for that doesn't take any appreciable ... proportion of the time to run the program. ... believe that multiple CPUs can speed up memcpy - what are you getting ...
    (uk.comp.sys.mac)
  • Re: RHEL high percentage of processor time from system?
    ... > The disk i/o is actually pretty low ... ... I run BOINC projects at nice 19 to gobble up ... > fortran vs. c vs. matlab with no disk i/o, ...
    (comp.os.linux.misc)
  • Re: Reading C code in matlab... mex file?
    ... directly from the matlab workspace. ... it seems that a mex file creates the matlab ... executable on the data file, then read the result back in. ... either the disk I/O is a small percentage of the processing time per ...
    (comp.soft-sys.matlab)