Processor Load and Util

[mtggit]

I have some questions on the zabbix agent on Linux and hope anyone can help me.

I have setup the agent on a Redhat ES3 server and have zabbix agent running. I am trying to monitor the CPU loading and so I looked at the system.cpu.load as well as the system.cpu.util.

1. I got the some data on the cpu load, but don't know how to interrupt. Most of the time the value is about 0.2-0.3 and the max value is just around 0.5. Any idea on the actual loading? What is the max value of the load?

2. I also looked at the cpu util, and the percentage is switching in a range of several percent, say 1-5. However, I use top to check the loading and the CPU is can reach about 30-40%. If the cpu util is an average load, is it possible to set the average of a shorter time period, say 5-10 seconds?

[RandySpinaGE]

1) A comment on question number one.

Many people get confused about cpu "load". What it is is the amount of processes waiting on the "run queue". So load is the amount of processes in the "I want to run pipe". Years ago I wrote a simple script to "while true; do : ;done. In otherwords do a noop forever. Pure CPU, no I/O. I was able to spawn enough processes to get the LOAD up to 88.0, yes 88 before HPUX somehow slam dunked the parent and killed them all.

So the limit of "load" is "it depends" - It depends on what the processes want to do and how much the load latency influences the real work need.

Today many systems can peak over 4 or 6 (briefly) and be ok, but if that were consistant, I'd look into faster (or different) architecture.

2) I don't know interval avg of cpu util - I always thought of it as instantaneous

[mtggit]

Thanks for the reply.

1. Seems the cpu.load is another representation of the loading of the system. I am still a bit pizzle why it can be a decimal figure if the definitation is no. of process.

2. Quite sure that the figure is somehow an average value as I can see the cpu util boost up to 30% for more than 5 seconds, but the value it returns is only 3%. However, what is the definitation of avg1, avg5 and avg15? The manual only said avg5 = average every 5 mins. Does it mean avg1 = average every min? If so, why the system poll the data every 5 second by default? Seems waisting CPU time.