Hi Jamal, Do you have a specific example? We have tried this on Intel (x86 and IA-64), AMD (x86 and x86_64) and PPC processors. Most show gains with none showing any detriment to performance with the test cases. The gains seen on the ones that show gains are significant in almost all cases.
If you have a specific case where you think that there are measurable problems with it we would be happy to hear it so we can take a look at it. Thanks. Cheers, John > -----Original Message----- > From: jamal [mailto:[EMAIL PROTECTED] > Sent: Thursday, December 01, 2005 7:41 AM > To: Kirsher, Jeffrey T > Cc: Jeff Garzik; netdev@vger.kernel.org; [EMAIL PROTECTED]; > Ronciak, John; Brandeburg, Jesse; Robert Olsson > Subject: Re: Resend [PATCH netdev-2.6 2/8] e1000: Performance > Enhancements > > > On Thu, 2005-01-12 at 02:36 -0800, Jeff Kirsher wrote: > > e1000 driver update > > > > Signed-off-by: Jeff Kirsher <[EMAIL PROTECTED]> > > Signed-off-by: John Ronciak <[EMAIL PROTECTED]> > > Signed-off-by: Jesse Brandeburg <[EMAIL PROTECTED]> > > > > 2. Performance Enhancements > > - aggressive prefetch of rx_desc and skb->data just like we > do for 10gig > > - align the prefetches to a dword to help speed them up > > - copybreak for packets < 256 bytes, ideally we would like > to modify ethtool to allow this value to be changed, helps > small MTU, many reassemblies case > > Hi there, > > Most of these performance improvements maybe machine > specific. i.e they > maybe detrimental to perfomance in other machines. It would be a good > idea to get test results from a wide ranging scan of hardware. > > Can you make them optional? hate to say this but ifdefs maybe one > approach. > > Example the copybreak may behave differently on opteron/NUMA than on > Xeons and the "excessive" prefetching is dependent on the > strides which > are machine specific etc. > There was a patch from David Morsberger many moons back which did > prefetch on skb->data that seemed to work on a wide array of hardware; > maybe you should revisit that and not be too ambitious and > just add the > one prefetch? > > cheers, > jamal > > - To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
