On Sun, Apr 17, 2016 at 10:24:32PM +0200, Samuel Thibault wrote: > Thanks for the investigation. I have added the attached patch to the > gnumach running on the buildds. That'll revert to the old page cache > policy, which seems to be reverting to the previous performance. > > Looking at ps -feMj, it seems that it's ext2fs which consumes much more > CPU time, thus increasing overall wallclock time. It'd probably be > interesting to profile ext2fs, to see what takes longer. Perhaps it's > the hash table which is less efficient since with the new page cache > policy there are much more cached files?
There are undeniably more scalability issues to tackle, but I strongly think we shouldn't revert the work that has been done. We'll keep working on those, and we now have nice numbers to compare with. I also think that, at some point, we should provide a 64-bits capable kernel caching interface, and write an ext4fs server. I'll see if I can spend some time on this after high memory. -- Richard Braun
