Re: tiobench

rwhron@earthlink.net
Thu, 23 May 2002 09:10:23 -0400


Andrew Morton wrote:
> tiobench sequential read throughput is enormously sensitive to the
> readahead setting. With the readahead at the default 128k
> with
>
> blockdev --setra 512 /dev/hda
>
> gives very nice throughput. But somehow 2.4.x seems to cheat
> and seems to do twice the readahead than the setting says, or something.

Bigger readahead helps dbench too.

Readahead in 2.5.x and dbench and tiobench sequential read
average throughput on my uniproc ide box was roughly:

kernel readahead dbench tiobench
2.5.0 - 2.5.7 8 8.1 mb/s 9.9 mb/s
2.5.8 - 2.5.10 0 3.5 mb/s 9.9 mb/s
2.5.11 - 2.5.17 256 13 mb/s 10.2 mb/s
2.5.8-akpm 1024 22 mb/s 18.1 mb/s

Bigger readahead appears to lower tiobench sequential read
latency a little.

2.4.x showed readahead = 8 for 2.4.3 - 2.4.19-pre8, including
aa, ac, jam, mjc, and rmap.

-- 
Randy Hron

- 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/