sar monitoring failure

On Monday we had a very strange problem on one of our servers. %iostat values reported by sar dramatically increased the values and stayed at that range. After long investigation I could not find any problem that could be causing such a havy IO load and then I noticed, that the output of iostat and sar differ for %iostat values.

It appears that older version of sar could have some kind of overflow problem that makes it show invalid data values after kind of long time after system startup.

Updating sar to the newer version helped this (or maybe just reset the counters somehow)