I'm wondering if anybody could shed any light on the behaviour of swapping priorities and what happens to paged state data that remains largely unreferenced.
I am tinkering with swap priority having had my Zaurus building the Ruby port for about 20 hours now. Basically I have a 128Mb swap partition on this Zaurus and to provide capacity whilst building large builds I have a 128Mb swap file. There is a significant performance difference between swapping to the file or to the partition and I noticed that based upon the default priorities (both devices as 0) the paging load was being evenly distributed between the devices as one might expect. Having used swapctl -c -p to change the swap priority on the swap file I am starting to see migration of load away from the page file to the swap partition and things have started to speed up. My question is really around unreferenced state data that has been pushed out to swap and isn't being demand paged back in. Is there functionality in the swap strategy to migrate such pages to a lower priority device so that you can bias performance of pages referenced more often against the higher priority swap device? If not then how can you maximise the benefits of the prioritisation mechanism to ensure the majority of the working set is on the highest priority partition and that the highest priority partition is not consumed on a first come first serve basis by unreferenced state data that falls outside of the general idle working set of processes. Any information is much appreciated, Regards, -Andy
