On Tue, 2002-04-30 at 11:53, Steve Lord wrote:
> On Tue, 2002-04-30 at 04:55, Steve Lord wrote:
> > Some multiple blocksize work, and better optimization of allocation
> > and writes in the mmap write case where memory pressure is the cause
> > of data getting flushed. We used to fragment files really badly in
> > this case, now fragmentation is reasonable - which causes major
> > speedup of some loads.
> >
>
> I would hold off taking this code until I do some more work on it, looks
> like I fixed one lot of performance by killing another one.... dbench
> throughput has gone down the tubes.
>
> Not sure yet when it came in - sometime in the last 24 hours I think.
>
> Steve
OK, I take that back, performance has not been killed by this, it was
my machine configuration on this end.
Steve
--
Steve Lord voice: +1-651-683-3511
Principal Engineer, Filesystem Software email: lord@xxxxxxx
|