Re: 2.5.69-mm1

[Dipankar Sarma <dipankar@in.ibm.com>]

On Tue, May 06, 2003 at 07:20:51AM -0700, David S. Miller wrote:
>    From: Dipankar Sarma <dipankar@in.ibm.com>
>    
>    Provided there isn't a very heavy contention among readers for the
>    spin_lock.
> 
> Even if there are thousands of readers trying to get the lock
> at the same time, unless your hold time is significant these
> readers will merely thrash the cache getting the rwlock_t.
> And then thrash it again to release the rwlock_t.

And now ISTR that this is indeed the case, atleast going by
what we saw with "chat" microbenchmarks (fwiw :)).
Hold times weren't very high and most of the performance penalty
came from bouncing of the rwlock cacheline, which prompted us to
write a RCU-based patch for lockfree lookup from fd table.

Thanks
Dipankar
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"aart@kvack.org"> aart@kvack.org </a>