Subject: Re: test kernel for cobalt
To: Izumi Tsutsui <tsutsui@ceres.dti.ne.jp>
From: Markus W Kilbinger <mk@kilbi.de>
List: port-cobalt
Date: 05/26/2006 20:14:25
>>>>> "Izumi" == Izumi Tsutsui <tsutsui@ceres.dti.ne.jp> writes:

    >> this additional gt.c patch seems to improve ata/ide disk
    >> performance (partially), but seems to decrease network (tlp0 in
    >> my case) performance.

    Izumi> Hmm. I've tested GENERIC kernels (updated today) with my
    Izumi> local patch, but I can't see network (tlp0) performance
    Izumi> decrease on my RaQ2: ---
    Izumi> [...]
    Izumi> Summary: ---
    Izumi> w/o gt.c change: ttcp-t: 16777216 bytes in 3.57 real
    Izumi> seconds = 4583.46 KB/sec +++ w/ gt.c change: ttcp-t:
    Izumi> 16777216 bytes in 3.44 real seconds = 4758.06 KB/sec +++

    Izumi> w/o gt.c change: ttcp-r: 16777216 bytes in 3.49 real
    Izumi> seconds = 4687.96 KB/sec +++ w/ gt.c change: ttcp-r:
    Izumi> 16777216 bytes in 3.47 real seconds = 4726.09 KB/sec +++
    Izumi> ---

You are right: I get similar results with a generic kernel on my
qube2, no performance decrease if your gt.c patch is activated.

The general decrease of network performance in my formerly reported
results seem to be related to the activated pf (especially due to
scrub rules).

    Izumi> If there is no objection, I'll commit these changes in a
    Izumi> few days.

So, no objection from this/my side ...

Markus.