On 12/7/05, Robert Olsson <[EMAIL PROTECTED]> wrote: > Thanks for this discussion. It comes in handy as I have to prepare > for a new distro for our network stuff. Right now it seems like > e1000 6.2.15 can be used but with HW-flow disabled, cpybrk disabled > and only with the two first prefetches. Hmm there were some patches > for using local memory on NUMA systems. What the status here? Tests?
The idea behind all of these tests have been to show if prefetch has _any_ case where it shows to be a detriment to performance over no using it at all. As far as I can tell there is no such case (looking at all the data generated so far). It's either been a gain (the first 2 prefetches in this case) or a no-op (the other prefetches). This is only with these test. Other test case show better gains with more normal traffic types (i.e. not all small packets). As far as copybreak goes, we knew it probably won't help routing type test with small packets. Robert's test shows it really only hurts where it seems to be CPU bound, which makes sense. This can be disable at compile time by setting E1000_CB_LENGHT to 2K which means that copybreak won't ever be done. Flow Control can be disable on a per interface case today so this isn't an issue either. There stills no case being shown which show a detriment to performance for prefetch (which was the goal of all these tests). We still don't see the case where it hurts. -- Cheers, John - 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
