Messages in this thread | | | Subject | Re: REGRESSION: Performance regressions from switching anon_vma->lock to mutex | From | Tim Chen <> | Date | Wed, 15 Jun 2011 14:12:16 -0700 |
| |
On Wed, 2011-06-15 at 13:57 -0700, Andi Kleen wrote: > > 7.44% exim [kernel.kallsyms] [k] format_decode > > | > > --- format_decod > > > This is a glibc issue. exim calls libdb and libdb asks sysconf for the > number of CPUs to tune > its locking, and glibc reads /proc/stat. And /proc/stat is incredible slow. > > I would blame glibc, but in this case it's really the kernel to blame > for not providing proper > interface. > > This was my motivation for the sysconf() syscall I submitted some time ago. > https://lkml.org/lkml/2011/5/13/455 > > Anyways a quick workaround is to use this LD_PRELOAD: > http://halobates.de/smallsrc/sysconf.c > But it's not 100% equivalent. >
Thanks to Andi for providing the info. We've used this workaround in our testing so it will not mask true kernel scaling bottlenecks.
Tim
| |