High idle load

[RoyK]

Hi all

Testing on two separate servers, one dedicated and one VM, I get a very high load, mostly system time, while running Zimbra. Both systems are mostly idle, only having around 5-10 users, 2-3 concurrent users at most. I don't see a very high load in 'top', but using sysstat/sar, I see a stable load of 40-60% while idle. Looking into it, it looks like java processes are started routinely into new processes, and I suspect they are started like shell scripts or the likes, having to be precompiled each time. I also see new PIDs coming up all the time in top. Someone first suspected this was a memory problem (1GB RAM), but the same happens with 2GB and still with memory unallocated.

Is this meant to be so, and if so, who on earth thought this was a good idea? It's quite some amount of CPU being wasted in these runs. If not, what can I do about it without moving to another system?

roy

[raj]

you did not tell what version and what os and other info, but maybe the following link will help you specially on VM install

High CPU spikes every 1 minute

* Please backup before trying any of the solution.

Raj

[RoyK]

Hi

Thanks for the quick reply

I'm running 6.0.3 (or was it 6.0.2?). the cron job is now changed, running every hour. But the zmlocalconfig/zmprov commands below doesn't seem to be right with v6. btw, what is the zmlogprocess?

roy

su - zimbra
zmlocalconfig -e zmmtaconfig_interval=6000
zmprov mcf zimbraLogRawLifetime 7d
zmprov mcf zimbraLogSummaryLifetime 30d
/opt/zimbra/libexec/zmlogprocess

[raj]

did you get any error when you run those in version 6.xx or you are asking without running them?

Raj

[RoyK]

At first I didn't try, since I didn't find any docs about what they did. Now, I did try nevertheless and I got no errors.

zimbra@zimbra:~$ zmlocalconfig -e zmmtaconfig_interval=6000
zimbra@zimbra:~$ zmprov mcf zimbraLogRawLifetime 7d
zimbra@zimbra:~$ zmprov mcf zimbraLogSummaryLifetime 30d
zimbra@zimbra:~$ /opt/zimbra/libexec/zmlogprocess
zimbra@zimbra:~$

Can you please explain what they do?

I'll get back to you on performance data

thanks

roy

[RoyK]

Seems this solved the problem, see sar output below, but still, what does each of these commands do? I like at least to have an illusion that I'm in control of my system

roy

20:25:01 CPU %user %nice %system %iowait %steal %idle
20:35:01 all 9.79 0.00 54.07 1.63 0.00 34.50
20:45:01 all 9.87 0.00 50.60 1.97 0.00 37.56
20:55:01 all 5.20 0.00 36.48 1.78 0.00 56.54
21:05:02 all 0.98 0.00 7.39 2.39 0.00 89.25
21:15:01 all 0.58 0.00 4.88 1.37 0.00 93.16

[ArcaneMagus]

I'll answer these from what I understand here...

zmlocalconfig -e zmmtaconfig_interval=6000

This sets the interval at which the zmmtaconfig process checks the postfix configuration files for changes and reloads them if there are any. This has been known to cause high CPU usage on some systems and is probably the only reason your CPU usage has dropped.

zmprov mcf zimbraLogRawLifetime 7d

This sets the maximum age for your log files, I'm not entierly sure what this would have to do with performance though as it just says when to delete the old log files. (7 days)

zmprov mcf zimbraLogSummaryLifetime 30d

Not entierly sure what this does as there is no "summary" log that I know of, but again from what I can find in the wiki this would only affect disk space usage, not CPU.

/opt/zimbra/libexec/zmlogprocess

From what I can see the only affect this would have is that it optimizes the logger tables when it runs, so it might have sped up access to them, thus slightly reducing CPU load on access to these tables.