performance problem PAE kernel

opeyrega · 09-28-2009, 11:11 AM

Hello

I've a VM ( esx 3.5 ) Linux server RH 5.2 32 bits with 8Go amount of memory

Quote:

Linux box 2.6.18-92.el5PAE #1 SMP Tue Apr 29 13:31:02 EDT 2008 i686 i686 i386 GNU/Linux

We've installed kernel-PAE-2.6.18-92.el5, but we have many memory problem performance.

The memory is correct

Quote:

]# free
total used free shared buffers cached
Mem: 7860632 5778616 2082016 0 31788 229172
-/+ buffers/cache: 5517656 2342976
Swap: 4192924 0 4192924

when we try the stress command on memory

Quote:

stress --vm 2 --vm-bytes 128M --timeout 20s

-> the system hangs, ntp is desynchronized etc ...
test performance on IO disk : ok
test performance on Cpu : ok

no log in /var/log/*; dmesg; syslogd of this memory problem.

Quote:

[root@slapp155 kernel]# vmstat 1 20
procs -----------memory---------- ---swap-- -----io---- --system-- -----cpu------
r b swpd free buff cache si so bi bo in cs us sy id wa st
0 0 0 2076840 35124 229904 0 0 11 3 255 40 1 1 98 0 0
0 0 0 2076716 35132 229896 0 0 4 47 1019 149 0 0 100 0 0
2 0 0 2019056 35132 229916 0 0 4 0 1056 149 0 33 67 0 0 # stress test on memory
2 0 0 1937216 35132 229920 0 0 0 0 1010 136 0 50 50 0 0 # stress test on memory
2 0 0 1874968 35136 229920 0 0 0 3 1213 136 0 50 49 0 0 # stress test on memory
0 1 0 2076592 35148 229920 0 0 16 26 1133 160 0 39 59 1 0
0 0 0 2076592 35160 229924 0 0 4 37 1035 147 0 1 99 0 0

We try to stop vmware-tool, used memory down to 2,5Go but still have the same bad performance.

no response via teacher google ....

maybe some value must be changed in the kernel ?

Any idea ( exept to re-install in 64bits ) ?

johnsfine · 09-28-2009, 12:17 PM

I don't think many people here will even understand the question you are asking, much less know the answer.

The fact that you are using esx 3.5 seems to be a key aspect of the problem, but you didn't include that in the subject line of your post. So experts who might know about esx might never look at your post.

You seem to have jumped to the conclusion that the PAE kernel is a significant factor in the problem. But I don't see anything in your post supporting that.

You also should use code tags rather than quote tags around text copied from the running system (especially that vmstat output) to make it readable for anyone who might take the time to help figure out your problem.

Maybe if I had experience with esx, I might be able to figure out what problem you are describing, but now I just see phrases like "memory problem" and "bad performance" and no specifics to say what you are even talking about.

I'm not sure what the stress command you are running stresses. I don't know what resources are provided from the real machine to the VM. I would guess this VM has 2 cores based on the fact that the stress test causes roughly 50% system time and 50% idle time.

If it is 2 cores, is 50% system time and 50% idle time unexpected? For example did you expect a larger fraction of time would be user? Or did you expect the kernel time would be multi threaded and thus make good use of both cores?