Matthew Kirkwood writes:
> On Wed, 19 Apr 2000, Alan Cox wrote:
>
> > > >From our measurement using notorious Mindcraft-like web-benchmark,
> > > Apache issues tremendous number of sys_newlstat() to check the
> > > requested path not including symbolic links. Which causes most of
> > > kernel functions locked-out by the lock.
> >
> > Apache isnt designed for speed, its designed for flexibility. Its
> > probably much more interesting to profile something like thttpd,
> > although that currently isnt threaded.
>
> You can get rid of all the lstats by doing:
>
> <Directory />
> Options FollowSymlinks
> </Directory>
Yes, we've already done with the setting and the performance becomes
better.
What I want to point out is not the apache ill-behavior, but the
kenrel behavior; the read-only operation, such as sys_stat, can
lock-out all-other kernel important activities. This becomes the
bottleneck in four or more CPUs system.
The most of the performance loss is caused by spinlock wait in the
stext_lock section. And precise observation shows that the
kernel_lock variable is guilty. This means longer locking duration by
parallel execution.
To gain the parallel execution speed-up, global lock should be avoided
as far as possible, it is needless to say..
-- Computer Systems Laboratory, Fujitsu Labs. kumon@flab.fujitsu.co.jp- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sun Apr 23 2000 - 21:00:15 EST