swappiness is the route to happiness

hmm.  Stuff like this is really hard for me to gauge, but it’s apparently hard for others too, as I browse through some of the old kernel mailing lists (circa like 2004).

I’ve been playing with different values in /proc/sys/vm/swappiness (it can be anywhere from 1-100).  The higher the number, the more a system will swap.

Mr. Andrew Morton sets his to 100, because he just ‘sticks his fingers in his ears and sings la la la until people tell him ‘I set the swappiness to zero and it didn’t do what I wanted it to do’.

He  says, “My point is that decreasing the tendency of the kernel to swap stuff out is wrong. You really don’t want hundreds of megabytes of BloatyApp’s untouched memory floating about in the machine. Get it out on the disk, use the memory for something useful.”

The other side is if ‘bloatyapp’ is swapped out too aggressively, when the user returns to it there is a delay.

Well, so far I set it to zero for a day and 100 for a day, going away from the default of 60.  To be honest, I’ve had a hard time gauging the performance, other than by feel.  I’ll have to think about some measurable benchmarks to look at normal lab usage on progeny.