Comments on: Using LoadAvg for Performance Optimization

By: Patrick Casey

Patrick Casey — Mon, 18 Dec 2006 16:44:04 +0000

I have to agree. If your flavor of unix has vmstat it’s infinately better than loadavg for performance tuning. I pretty much always have a terminal up running vmstat 5 on my servers.

By: Xaprb

Xaprb — Wed, 06 Dec 2006 18:57:31 +0000

I agree with Eric. I have never been able to make anything meaningful out of loadavg. I had to dig into the kernel source code to even learn what those numbers came from, and that still didn’t help me understand why they were a meaningful metric of the system’s real load.

However: they are great to track over time how much load the box is under. I can’t use them for comparing load on different boxes very well, but it does help me to see how the box is doing compared to itself last week (or last hour). I like to run tload on some servers every now and then. But generally I just run vmstat -n 5.

By: peter

peter — Wed, 06 Dec 2006 09:30:45 +0000

Yes I agree. VMSTAT/IOSTAT are better to look at.

I just simply find many people having trouble understanding vmstat/iostat – there are just way too many numbers while loadavg is simple 3 numbers which they try to judge load by.

By: Eric Bergen

Eric Bergen — Wed, 06 Dec 2006 05:50:55 +0000

For most of the reasons above I completely ignore load average. I honestly wish that the tool maintainers would remove it and distros would replace it with a decent sar config. Load average can be affected by too many variables (process count, cpu count) to make it easily comparible. In some cases a load average of 5 is bad and in others 50 is ok. It’s very difficult to read without knowledge of the system and what’s running on it.

With vmstat and iostat you can see a clear percentage value of how busy something is. When the counter gets to 100% it’s 100% busy. Simple as that.