When should I alert ?

[KapilRaj]

Guys,

I have a basic doubt / requirement and here it goes ..

I believe CPU idle time "zero" is not a problem which needs an immediate fix. 'Cause if CPU is free, programs will use it. But what if the runq is also more ? , yes there is a problem.

Can I have a formula for an alert script using vmstat's runq ,idle time and number of processors ?.

i.e. If I have 4 processors and the runq is "50" whether it is a problem or not ?. And what if I have a single processor and a runq of "50"

I can not go for special tools like glance, perfview or something like that.

Regds,

Kaps

[Mark Grant]

I have to admit that I go for load average when I can't think of a specific thing to monitor. My reasoning is that if the load average is high, something is wrong. It doesn't tell me much else but I can find that out myself when I have a look. load average catches all sorts of performance problems.

[Ralph Grothe]

Hm, as you only mention vmstat I guess you cannot get hold of glance, or the measureware agents?
If you had it installed you would find the file /var/opt/perf/alarmdefs
which gives a pretty sound alarm definition for an indication of a CPU bottleneck,
that you could use as a guideline.
Of course, it uses MWA metrics, but their names are pretty self-explanatory.

Here's the CPU bottleneck definition from that file

symptom CPU_Bottleneck type=CPU
rule GBL_CPU_TOTAL_UTIL > 75 prob 25
rule GBL_CPU_TOTAL_UTIL > 85 prob 25
rule GBL_CPU_TOTAL_UTIL > 90 prob 25
rule GBL_PRI_QUEUE > 3 prob 25

alarm CPU_Bottleneck > 50 for 5 minutes
type = "CPU"
start
if CPU_Bottleneck > 90 then
red alert "CPU Bottleneck probability= ", CPU_Bottleneck, "%"
else
yellow alert "CPU Bottleneck probability= ", CPU_Bottleneck, "%"
repeat every 10 minutes
if CPU_Bottleneck > 90 then
red alert "CPU Bottleneck probability= ", CPU_Bottleneck, "%"
else
yellow alert "CPU Bottleneck probability= ", CPU_Bottleneck, "%"
end
reset alert "End of CPU Bottleneck Alert"

[KapilRaj]

Ralph,

Hmmm ... i never worked on perf / glance . Can u explain in detail please ?.

Kaps

[Naveej.K.A]

hi kapil,

You get a 60 days evaluation version of glance in one of the applications CD supplied with the server. you can install and have a GO at it...

If you do have an HP9000 box!!!!

with best wishes
Naveej

[KapilRaj]

In an SMP System ,my current monitoring algorithm is to alert when,

01.runq is more than 3 times the number of processors
02. If wait is more than 25 %

Can anybody advise if this needs improvement / any suggessions are welcome.

Regds,

Kaps