How to tell when a job is swapping?
Daniel Kidger
Daniel.Kidger at quadrics.com
Tue Feb 19 06:55:33 PST 2002
-----Original Message-----
From: Jeff Layton [mailto:jeffrey.b.layton at lmco.com]
> I usually assume that if the free swap space falls below
>the maximum and load starts climbing that the node is
>swapping. However, when I talk to the user, he states that
>the code is running fine and the timing numbers are where
>they should be. So, I'm obviously interpreting something
>incorrectly (unless the job is really swapping but for some
>reason performance is unaffected).
if a job allocates large areas of memory (beyond physical available) , uses
them once - say fills them with zeroes and then doesn't use much of that
array space, then:
a/ you will be using swap space
but
b/ performance is unaffected since these (unused) pages will get swapped
out pretty quickly, but never get swapped back in since they are never
referenced again.
If you want to watch swapping actually happening then one way is to run
vmstat, e.g. "/usr/bin/vmstat 2". look at the si and so columns.
eg. I run 2 copies of a code that almost uses up all memory and then I
start a third copy (last two lines below) which causes the machine to start
swaping frantically ...
[root at dell7 /root]# vmstat 2
procs memory swap io system
cpu
r b w swpd free buff cache si so bi bo in cs us sy
id
2 0 5 29540 620624 1688 7876 49 81 201 84 71 30 84 1
15
2 0 5 29540 620624 1688 7876 0 0 0 0 127 8 100 0
0
2 0 5 29540 620624 1688 7876 0 0 0 0 127 10 100 0
0
2 0 5 29540 620624 1688 7876 0 0 0 0 128 8 100 0
0
3 0 5 29540 269312 1688 7876 0 0 0 0 139 38 73 27
0
3 0 5 29540 43816 1688 7876 36 0 36 0 135 32 84 16
0
3 0 5 73612 3356 1672 7788 0 20732 0 20778 214 718 82
18 0
2 1 5 115668 5700 292 4668 80 22100 80 22100 541 1468 78
22 0
Yours,
Daniel.
--------------------------------------------------------------
Dr. Dan Kidger, Quadrics Ltd. daniel.kidger at quadrics.com
One Bridewell St., Bristol, BS1 2AA, UK 0117 915 5505
----------------------- www.quadrics.com --------------------
More information about the Beowulf
mailing list