Auto Tuning Thresholds

[shohn]

Okay, so I've been trying to figure out a way to only send out an alarm upon exceeding a system's normal load levels as averaged out over the last week or month for example.

Has anyone had any experience tips to alarm on some key metrics to focus in on that would show when a system has gone beyond an established baseline - perhaps even on an hourly basis?



I tried alarming if the minimum load exceeds the average load over the last week and it generates a ton of noise.

[nelsonab]

If you don't mind a little hacking I've written a little ruby script that can get you part way there. I put it in the Wiki
http://www.zabbix.com/wiki/doku.php?id=contrib:scripts

All you need to do is tell it an itemid and it will return the average of that item for the last hour, even if the item is stored as a delta. If you run the script with no command line options you will see a help prompt. It requires the MySQL Ruby libraries.

If you were to write a shell script that takes this number and puts it into the expression field of your trigger along with all the other pertinent information you would then have a "dynamic" trigger.

I'll see if I can't write a script to handle that too in the next few days.... :-)

[shohn]

Averg - Std Dev

So am I not understanding how the average function works under triggers? It seems like it would be something as simple as alarming only if


Perhaps a std deviation function or something could be included, thereby I could set alarms only if the minimum is greater than the average and perhaps two or three standard deviations for a set period of tiem to help minimize the outliers.

I wonder how easy it would be to plugin a std deviation function into the zabbix server code.

Any ideas?