Comments on: Load Average

By: etbe

etbe — Tue, 06 Nov 2007 21:17:24 +0000

Peter: Good point. I will add an item to my todo list to write more about this.

By: Peter Moulder

Peter Moulder — Tue, 06 Nov 2007 13:15:39 +0000

Incidentally, the traditional description of “average over the last 1, 5 and 15 minutes” is really a piece of fiction, at least regarding Linux. They’re exponentially smoothed values, and they don’t really correspond to “1 minute” or whatever in any meaningful way that I know of; e.g. the smoothing values chosen don’t minimize the sum of (absolute or squared or cubed or ^1.1) differences between the exponentially smoothed value and the true 1 minute (etc.) moving average. The only meaningfulness that I know of for the expression used to calculate the smoothing value in Linux (viz. 1/exp((update interval)/{1,5,15}min)) is that at least the ratio 1:5:15 is meaningful, even if the absolute time durations aren’t. I’d be interested to hear if someone knows why that expression was chosen, or why it’s useful to use that expression rather than one that gives a value closer to the 1min moving average.

By: etbe

etbe — Fri, 02 Nov 2007 21:21:11 +0000

The disks in question are a year old and were not particularly fast when I bought them. Doubling the performance is quite possible.

iostat does display what it considers to be the disk usage percentage, not sure how accurate it is – I know that the 0% and 100% values are right though. ;)

By: Olaf van der Spek

Olaf van der Spek — Fri, 02 Nov 2007 12:15:42 +0000

BTW, why is there no simple disk usage metric? Like, this disk is busy 50% of the time.

By: Olaf van der Spek

Olaf van der Spek — Fri, 02 Nov 2007 12:13:07 +0000

> But if the disk IO performance was doubled (a realistic possibility given that the system has a pair of cheap SATA disks in a RAID-1) then find would probably use 8% of CPU time.

Doubling STR does still not (always) double performance. ;)