Garrett Cooper <yaneg...@gmail.com> wrote: > It would be interesting to see how well FreeBSD fares now because > 7.1 was before the ULE scheduler was setup as the default scheduler. <...>
We are well aware of ULE and tried with it enabled as well. :) > <...> Things also improved > again with scaling in FreeBSD 8.x, and there's work in the pipeline > for 9.0-RELEASE to get rid of vm-locking which seems to have > qualitatively yielded performance gains on the 2 workstations and 1 > server I have FreeBSD CURRENT installed on and actively work with. Not sure what "vm-locking" you have in mind. I have seen some commits which reduce contention on page-queue lock by moving protection of some VM page flags to an object's lock. This issue is relevant for NetBSD as well. We investigate other approaches, though. > If I can get things up and running like someone did for the NetBSD > slides above with FreeBSD CURRENT and NetBSD CURRENT on my amd64 test > machine (directions welcome :)..), I'll give the tests a shot again to > see how divergent the performance results are between the two BSDs in > this area. Feel free to post them on tech-perform@, it will be interesting to know. Thanks. -- Mindaugas
