On Wed, May 23, 2012 at 1:05 PM, Maksim Yevmenkin < maksim.yevmenkin_at_gmail.com> wrote: > hello, > > would anyone object to the following patch? > > No objection. There shouldn't be any controversy here. Your patch is correct. The existing code in UMA is doing the wrong comparison. Alan > === > > Index: uma_core.c > =================================================================== > --- uma_core.c (revision 616) > +++ uma_core.c (working copy) > _at__at_ -267,10 +267,7 _at__at_ > static void > bucket_enable(void) > { > - if (cnt.v_free_count < cnt.v_free_min) > - bucketdisable = 1; > - else > - bucketdisable = 0; > + bucketdisable = vm_page_count_min(); > } > > === > > i've observed situation where per-cpu buckets were disabled while > there were enough free cached pages. basically, cnt.v_free_count was > sitting stable at a value lower than cnt.v_free_min and that caused > massive performance drop. tuning down vm.v_free_min sysctl immediately > helped. > > thanks, > max >Received on Wed May 23 2012 - 16:15:52 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:27 UTC