Critical Load calculation.

[Prakash Nair]

I am collecting data using sar, uptime, vmstat, glance, top ...
How will I know at what load the system will crash or how will I arrive at a figure beyond which the system might crash?
For example if load average is above 3 the system will crash, is their any way to arrive this figure based on the system configuration.

Thanks
Prakash

[Rick Garland]

Crashing is not usually an event because of load. You will probably have diminished response and long periods of waiting because the box is so busy. The load of a system is dependent on many factors and having a system with a load of 1 vs a system with a load of 10 is not necessarily a bad thing. The system with the load of 1 could be having more trouble. Look at the reponse times - if the users are happy with the response on the system with a load of 10, the system appears to be doing its job. BTW, I have seen loads in the upper 90s and the system did not crash. Very slow, but did not crash.

[Brian M. Fisher]

The simple response is the system should NOT crash at any load. We are talking about UNIX not NT.

Brian
<*(((>< er

[Tim Malnati]

As some others have stated, this is somewhat of an individual thing based on the environment. A good thumb rule would be to watch the system closely if the load average get above 10. I have seen some systems get considerably above this and take care of themselves. But with high loading there is a good chance that you can can get to the point where heavy thrashing will take place to the extent where you may never return without killing off processes. This is particularly problematic if there is a high level of swap activity. If your load average goes above 100, odds are your system is crawling. In this situation it will probably be difficult to find the processes that need to be killed off where the diagnostic processes will be slow as well. In most of these situations the only realistic thing to do is to reboot.

[Devbinder Singh Marway]

Another little thing to note ( probably done already) is the specification of the server . This depends on the number of users on the system and the application running on the server. The common problem is ' my machine is running slow' , when doing sar, vmstat etc.. , it may show cpu bound , ( there again it may just be the application depending on what it is doing i.e. large queries etc..)

I have seen , when a machine crashes , the first response is users complain about the speed , ( network is checked out and is fine), you check system status and find cpu bound and memory intensive 0 the normal stuff
and app is hammering the box , another thing to look for is the streams whether there is enough available , because this will cause server to a halt if not enough are set.

This is a very 'big' area , and everything needs to be looked at i.e. server, appliction,number of users, kernel parmAMETERS, network , )

I hope this helps