Hi,
We have a problem with our Alphastation 400/4/233 that when we run a
certain fortran code, we only get 20% of cputime even when there is no one
or no job running other than our fortran program. The same program can get
about 98% cpu time on other DEC Alpha Jensen 200.
On the alphastation that we have problem (Alphastation 400/4/233),
"ps aux" command output:
USER PID %CPU %MEM VSZ RSS TTY S STARTED TIME COMMAND
lihsin 925 21.0 16.6 55.7M 217 ttyp4 R + 10:27:47 0:13.78 mgal
root 0 0.0 7.5 263M 9.5M ?? R < 13:59:43 0:05.48 [kernel
root 17 0.0 0.0 184K 32K ?? S 13:59:51 0:14.01 /sbin/up
root 395 0.0 0.2 1.39M 224K ?? S 14:00:13 0:00.20 /usr/sbi
"top" command output from the Alphastation 400/4/233:
load averages: 0.53, 0.46, 0.34 10:43:19
58 processes: 1 running, 1 waiting, 22 sleeping, 33 idle, 1 stopped
Cpu states: 13.9% user, 0.0% nice, 12.3% system, 73.7% idle
Memory: Real: 84M/118M act/tot Virtual: 47M/499M use/tot Free: 24M
PID USERNAME PRI NICE SIZE RES STATE TIME CPU COMMAND
925 lihsin 42 0 57M 22M WAIT 0:12 20.80% mgal <<==== problem
526 hartmann 44 0 11M 2285K sleep 0:31 1.20% dxterm
458 root 42 -2 22M 5193K sleep 1:21 0.90% Xdec
937 lihsin 44 0 2776K 401K run 0:00 0.40% top
842 hartmann 44 0 2744K 581K sleep 0:02 0.30% vi
499 hartmann 44 0 2792K 491K sleep 0:11 0.00% fvwm
the "mgal" job started with about 90% performance, than it decreases and
stays at about 20% cpu used. The job is in WAIT state most of the time.
While on the other alpha (Jensen 200MHz), "ps aux" command output:
USER PID %CPU %MEM VSZ RSS TTY S STARTED TIME COMMAND
lihsin 28871 31.0 14.6 54.9M 119 ttyp1 R N May 20 19:00:48 mgal
pmilne 24081 31.0 0.4 2.96M 336K ?? R N May 12 3-13:24:57 snw7n
jbrown 29477 31.0 1.6 5.66M 1.3M ?? R N 21:06:13 07:31:28 jason.ex
lihsin 24575 1.0 8.9 15.7M 6.9M ?? S May 13 12:02.34 /usr/loc
root 460 1.0 7.1 10.3M 5.6M ?? S < Apr 28 56:43.95 /usr/bin
lihsin 21803 0.0 0.5 2.73M 384K ?? S May 06 2:37.37 /usr/bin
root 21773 0.0 1.5 10.7M 1.2M ?? S May 06 2:18.88 /usr/bin
root 20922 0.0 1.4 9.96M 1.1M ?? S May 06 4:13.71 dxterm -
On this Alpha Jensen 200MHz we see the "mgal" can have 100% performance
if it is the only job and becomes 31% as it shares cpu with other two jobs.
On the alphastation where we have problem, we just installed another
64MB RAM to make it has a total of 128MB RAM and this alphastation
has 500MB swapspace. We reconfigured the kernel after the additional
RAM installation and the increase of swapspace. The additional RAM is
from CAMINTONN P/N 545774.
While in the other Alphastation (Jensen 200) it has 80MB RAM and
has a total of 400MB swapspace
Could you tell me how I can get the performance that it should be?
I run other fortran programs and I can get 100% performance on both machines,
only for this big memory usage code ("mgal") the results are different in cpu
usage. Actually the Alphastation which has the problem has larger RAM and
swapspace than the other alpha that gives 100% performance on the code.
Is this problem related to cache and slow memory chips?
Thanks,
lihsin
Received on Sun May 26 1996 - 18:47:55 NZST