Figure 4-15 is a vmstat report showing zero for both pages swapped in (si) and pages swapped out (so). Just as we want, the swap-out value is consistently zero. If you occasionally see pages being swapped out, that is an indication the system is under some memory pressure. It doesn't necessarily mean there is a problem, but it's getting close. I look at it as if the system is just entering the response-time curve's elbow.

I am commonly asked why I don't look at the pages swapped in and only the pages swapped out. Well, if there are pages being swapped in, there must have been pages swapped out. And if pages have been swapped out, there is a good chance they will eventually be paged back in.

Figure 4-15. A vmstat report on a very active four-CPU core database server. The so column represents the number of pages swapped out per second during the reporting interval. As in this sample, we want to this nearly always to be zero.