On Monday 15 July 2002 07:30, Andrew Morton wrote:
> Lincoln Dale wrote:
> >
> > Andrew Morton wanted me to do some benchmarking of large files on ext2
> > filesystems rather than the usual block-device testing
> > i've had some time to do this, here are the results.
> >
> > one-line summary is that some results are better, some are worse; CPU
> > usage is better in 2.5.25, but thoughput is sometimes worse.
>
> Well thanks for doing this. All rather strange though.
One result that seems pretty consistent in these tests is that avoiding the
page cache is good for about 20% overall throughput improvement. Which is
significant, but less than I would have thought if bus bandwidth is the only
major bottleneck. Something in the vfs/filesystem/blockio path is still
eating too much cpu.
Another observation: though only one of the tests hit 100% CPU, total
throughput is still shows consistent improvement as a result of reducing CPU.
This should not be, it means there is excessive latency between submission of
requests, that is, the IO pipes are not being kept full.
-- Daniel - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue Jul 23 2002 - 22:00:24 EST