| From: |
| Judith Lebzelter <judith@osdl.org> |
| To: |
| <suparna@in.ibm.com> |
| Subject: |
| OSDL: AIO-Stress summary for AIO patch set vs linux-2.6.8 |
| Date: |
| Fri, 27 Aug 2004 16:16:43 -0700 (PDT) |
| Cc: |
| <linux-aio@kvack.org>, <mason@suse.com>, <linux-osdl@osdl.org>,
<linux-kernel@vger.kernel.org> |
Hello;
I have done a summary of some aio-stress runs on our 1CPU and 2CPU hosts
to compare the performance of the latest version of Suparna's buffered AIO
patch set http://marc.theaimsgroup.com/?l=linux-aio&m=109285137...
(less the aio-ra) to the linux-2.6.8 kernel. I found up to a 442%
improvement and was wondering if anyone could explain it.
The buffered results on 1CPU/single disk/512M RAM/1 2G File showed:
Sequential Reads: -13.2%
Random reads: +32.7%
The buffered results on 2CPU/5 Disk striped Megaraid/1G RAM/1 4G File
showed:
Sequential reads: +2.1%
Random Reads: +442%
The improvement in buffered random reads from this patch set was +32% on
1CPU and +442% on 2CPU. This was more than expected on the 2CPU host at
+442%. It seems as though the 'buffered random reads' performance on the
unpatched linux-2.6.8 kernel is unexpectedly low at only 7.57MB/s on our
5 Disk striped Megaraid partition.
The buffered sequential reads on the 1CPU host were 13% slower, but they
did not change significantly on the 2CPU host.
There were no other serious regressions or changes within error.
The full summary is at:
http://developer.osdl.org/judith/aio/result_compare_2.6.8...
Regards;
Judith Lebzelter
OSDL
-
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/