cvs commit: src/share/man/man9 locking.9 rmlock.9 src/sys/conf files src/sys/kern kern_rmlock.c subr_lock.c subr_pcpu.c subr_smp.c src/sys/sys _rmlock.h lock.h pcpu.h rmlock.h smp.h

[Robert Watson]

On Sun, 25 Nov 2007, Alfred Perlstein wrote:

> * Julian Elischer <julian at elischer.org> [071125 10:05] wrote:
>>
>> not sure why sx-locks exist at all, as they seem to be a variant of sleep. 
>> I think it's just a convenience function set to allow one to implement a 
>> sleep-derived synchronisation.
>
> You are correct, sx locks are a faster replacement for hand rolled msleep 
> locks.  They're not only easier to read, but the underlying implementation 
> is faster.

We shouldn't under-emphasize the performance point: sx(9) locks instead of a 
hand-rolled msleep(9) lock for struct filedesc are one of the most important 
performance improvements in 7.0, and responsible for a large part of the MySQL 
improvement.  A big part of the improvement is avoiding nasty thundering herd 
behavior when there's contention.  Likewise, moving to sx(9) instead of 
hand-rolled msleep(9) locks in the socket buffer code made a noticeable 
difference in performance as well, although the thundering herd there was much 
less of a problem.

Robert N M Watson
Computer Laboratory
University of Cambridge