An otherwise good article recently in Linux magazine contained the following:
However, like too many similar articles, it fails to say what a high ratio of context switches, or too high interrupt rate, would be. On a Pentium dual core laptop, I'm seeing 100 to 1,000 cs's and around 60 - 800 interrupts per second (but often zero running processes), just surfing the web, while an old PC that I'm currently flogging by making it compile gcc is showing less than 10 context switches, but more than 300 interrupts.

Obviously, these machines are doing very different jobs. The old PC isn't having to run a GUI, and doesn't have a user prodding it constantly. However, is there a table or a formula to tell me "how high is too high"?

Thanks,
Rob

Hi -

The correct answer is:You need to:
1. Establish a performance baseline
2. Determine what factors .. on your particular system ... affect performance (one way or the other)
3. Decide what should be done

For example:
1. Let's assume your system gets "slow" when you run a particular app
2. Is the app CPU bound? I/O bound? Memory bound?
3. These are precisely the questions you need "vmstat" and "sar" to help answer
4. Whether you buy a new CPU, buy more RAM, tune your app or just live with the problem: that's completely up to you

'Hope that helps .. PSM

I was wondering whether the answer might be something like that. I suppose a high load for a single core machine running at 166MHz and a dual core one running at 10 times the clock rate is quite different.

Thanks,
Rob

Hi -

Yes, in general a 2GHz CPU should perform 100x the work of a 200MHz CPU, or a dual-core about twice the work of a single core.

However, if an app has code like this, you're always going to get 100% CPU:In any case, the answer is "it depends..."