Date: Thu, 17 Sep 2009 10:08:49 -0700 From: Pyun YongHyeon <pyunyh@gmail.com> To: Norikatsu Shigemura <nork@freebsd.org> Cc: freebsd-usb@freebsd.org Subject: Re: CFT: axe(4) performance patch Message-ID: <20090917170849.GL1155@michelle.cdnetworks.com> In-Reply-To: <20090918000657.099df7c4.nork@FreeBSD.org> References: <20090914193722.GD1155@michelle.cdnetworks.com> <20090918000657.099df7c4.nork@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Sep 18, 2009 at 12:06:57AM +0900, Norikatsu Shigemura wrote: > Hi pyun! > > On Mon, 14 Sep 2009 12:37:22 -0700 > Pyun YongHyeon <pyunyh@gmail.com> wrote: > > I submitted axe(4) performance patch to Hans and he committed the > > patch to P4. > > http://perforce.freebsd.org/chv.cgi?CH=168457 > > I tested following environments: > Thanks a lot nork@ ! > AX88178(GbE), AX88172(100), AX88772(100), bge(GbE), re(GbE), rl(100) > > patched: AX88178, AX88172, AX88772 <-> bge > old: AX88178, AX88172, AX88772 <-> bge > patched: AX88178, AX88172, AX88772 <-> rl > old: AX88178, AX88172, AX88772 <-> re > > My environment, there are 3 machines: > +--+----+-------------+----+ (All port GbE L2SW) > | | | > bge re rl > (patched) (old) > > bge: FreeBSD 7.2-stable > re: FreeBSD 9-current > rl: FreeBSD 8.0-beta4 > > > For test environments: > +--+----+---+---+---+ +--+-----+--+------+---+ > | | | | | | > bge re rl-axe bge re-axe rl > (old) (patched) > > > According to netperf, > > patched: AX88178, AX88172, AX88772 <-> bge > 60Mbps, 60Mbps, 25Mbps > patched: AX88178, AX88172, AX88772 <-> rl > 90Mbps, 60Mbps, 40Mbps > > old: AX88178, AX88172, AX88772 <-> bge > 180Mbps, 90Mbps, 95Mbps > old: AX88178, AX88172, AX88772 <-> re > 180Mbps, 90Mbps, 95Mbps > I'm not sure I understood the test environment. But it looks un-patched axe(4) performs better, right? One odd thing is performance differences for AX88172. My patch does not touch AX88172 controller but your benchmark shows big differences for AX88172. Are you sure you didn't apply other patches for axe(4)/USB stack? Ehh, I was wrong, there was a change for AX88172. The bufsize was changed to 16KB from 2KB for AX88172, that was not my intention. It seems new USB stack has no easy way to configure this parameter in attach phase so I used 16KB. Would you try changing the value to MCLBYTES(aorund line number 208 in patched if_axe.c) and test it on AX88172? Also please let me know what netperf parameters were used in the test. > Ummmm, I'll try to update old(rl) machine to pached, and re-test. > > Thank you. It seems you have all three variants that axe(4) supports, so would you test "http://p4db.freebsd.org/chv.cgi?CH=168602"?
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20090917170849.GL1155>