On Tue, 2005-06-12 at 16:55 +0100, Robert Olsson wrote: > Ronciak, John writes: > > > So we still need to see a case where performance is hurt by the > > prefetching. We have some data coming from another group here at Intel > > next week which we'll share once we have it which also shows the > > performance gains with prefetching. > > Hello! > > Well here is another small test with prefetching in 6.2.15. Workload is > still forwarding but now forcing both hash spinning, FIB lookups and GC > to run as 32k concurrent flows at 10 packets each are injected. Routing > table 123 kroutes 64 byte pkts. Dual Opteron 1.6 GHz with dual 82546GB > at 100 resp. 133 MHz. Aggergated performance from both CPU's. > > 0 prefetch 756 kpps > 1 prefetch 805 kpps (first prefetch only) > 2 prefetch 821 kpps (first two) > 5 prefetch 803 kpps (all) > > As seen this indicates an optimum with the two first prefetches in ths > setup. Seems I should have tested 3 and 4 as well... >
Nice work Robert! I am going to try and validate your results as well on one of my newer systems. cheers, jamal PS:- to Grant: as i found out, the case #3 of issuing a prefetch that is late depends on the architecture. On ARM a pre-load for example is a "hint" that the referenced address loading is needed. Therefore the core system will remember. - 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