iowait - Houston we have a problem - or?
The metric iowait is approximately as cumbersome as the memory free in unix. From time to time there is a major issue with the servers not coping with the demands and the applications see high iowaits. I always stumble when I try to explain that the metric doesn't necessarily tell you the truth. It seems like everyone knows that iowait is the time spent by the CPU waiting for IO to complete and that sounds bad. Back in the days it was bad, there was one CPU and it couldn't do anything until the IO completed. But now a days we have more core's i.e. other processes can continue on core's that aren't blocked by the single IO-wait. To add further insult the increase of CPU-performance has outperformed the improvements of disk performance. SSD disks are still fairly expensive, at least for larger disks event though they do stand for a gigantic leap in performance improvements. This means that we have to factors (CPU speed and multi-cores) that mitigate the issue o...