Keir Fraser wrote:
On 10/9/07 13:03, "Ben Guthro" <[EMAIL PROTECTED]> wrote:

I'm inclined not to backport this one.

If I recall - It applied against our 3.1 tree without any backporting...we just exported, and applied it. It increased performance on Caneland machines greatly. Test results against our 3.1 based product below:

<test results>
At Ben's request, I did a quick evaluation of the APIC TPR patch for Caneland. I used yesterday's build to establish a baseline for booting, running SPECjbb2005, and netperf on a SMP XP guest. I then repeated the tests with a custom kernel. The patch showed significant improvement for 2 of the 3 tests I used. Here are the results:

Test                 20070816     Patch    % Improvement
Boot time - Seconds    62.6        40.5         35%
SPECjbb2005 OPs/Sec   35216       35686          1%
TCP XMIT (MBits/sec)   70.2       309.5        341%
TCP RCV (MBits/Sec)   122.3       423.5        246%

These tests were done on a Caneland, with 4 quad-core sockets and 32GB of memory. The guest is Windows XP Professional with SP2, 2 CPUs, 2GB memory. This afternoon, I'll repeat the experiment on a non-Caneland machine to see if there are any
side effects.

</test results>

The two Linux changesets are not applicable to 3.1.
Yes, of mistake. I forgot to weed out my "unstable-only" patches from the list.

Xen-ppc-devel mailing list

Reply via email to