Ad Widget

**Aaron** · 03-07-2006, 18:31

I have a similar, but different problem on my Solaris 9 servers. My idle,user,kernel,wait data keeps increasing slightly each time it is sent. If I look at a 1 day graph, all 4 appear to be straigh lines. If I check the 1 week graph, I can see they all angle up slightly. The data I get from zabbix seems to have no realtion to the actual usage of the server.

I have been searching the forums for posts on system.cou.util and it seems there are several questions, but very few answers.

**Aaron** · 05-07-2006, 23:05

I am pretty sure I know what the problem with this is. I don't know how zabbix collects this data, but it appears to be the same data you get from mpstat. mpstat returns this data as an average from the time of the last reboot. Basically, the longer your server has been up, the less this data changes.

I haven't written C code since 1998 so I did not even attempt to verify this in the code. The numerical corelation seems to be spot on though so this is probably correct.

I was toying with a UserParameter hack for this, but it seems like the sort of thing that really should be fixed in zabbix. The data the agent is collecting seems usless in my opinion.

The hack I am writing is basically a cron that collects mpstat data and dumps it to a /tmp file that zabbix can just cat/sed/awk quickly. I haven't worked out the period of the collection or the check. It seems like there should be a better way, but I haven't thought of one yet.

**art** · 05-10-2006, 06:54

Had a quick look at zabbix-1.1.2/src/libs/zbxsysinfo/solaris/cpu.c

Looks like current SYSTEM_CPU_UTIL function uses kstat system calls which returns cpu utilisation counters since last reboot(?) - therefore zabbix shows average cpu utilisation for quite long period of time instead of current state of things. Though I'm not a C programmer (well, not programmer at all) I patched cpu.c so it now returns more meaningful data ( more or less the same as sar does).

Again, I'm not programmer and possibly there is more correct way to work with kstat data but it works for us ( tested on Solaris9 ).

Patch:

79a80,81
> unsigned long long cpu_val_s[4];
> unsigned long long cpu_val_e[4];
140c142
< if (get_cpu_data(&cpu_val[CPU_I], &cpu_val[CPU_K], &cpu_val[CPU_U], &cpu_val[CPU_W]))
---
> if (get_cpu_data(&cpu_val_s[CPU_I], &cpu_val_s[CPU_K], &cpu_val_s[CPU_U], &cpu_val_s[CPU_W]))
141a144,149
> sleep(5);
> int q = get_cpu_data(&cpu_val_e[CPU_I], &cpu_val_e[CPU_K], &cpu_val_e[CPU_U], &cpu_val_e[CPU_W]);
> cpu_val[CPU_I] = cpu_val_e[CPU_I] - cpu_val_s[CPU_I];
> cpu_val[CPU_K] = cpu_val_e[CPU_K] - cpu_val_s[CPU_K];
> cpu_val[CPU_U] = cpu_val_e[CPU_U] - cpu_val_s[CPU_U];
> cpu_val[CPU_W] = cpu_val_e[CPU_W] - cpu_val_s[CPU_W];

Ad Widget

system.cpu.util on solaris

system.cpu.util on solaris

Comment

Comment

Comment