On Sat, Apr 15, 2017 at 09:28:33PM +0200, Jesper Dangaard Brouer wrote: > On Sat, 15 Apr 2017 15:53:50 +0100 > Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx> wrote: > > > This reverts commit 374ad05ab64d696303cec5cc8ec3a65d457b7b1c. While the > > patch worked great for userspace allocations, the fact that softirq loses > > the per-cpu allocator caused problems. It needs to be redone taking into > > account that a separate list is needed for hard/soft IRQs or alternatively > > find a cheap way of detecting reentry due to an interrupt. Both are possible > > but sufficiently tricky that it shouldn't be rushed. Jesper had one method > > for allowing softirqs but reported that the cost was high enough that it > > performed similarly to a plain revert. His figures for netperf TCP_STREAM > > were as follows > > > > Baseline v4.10.0 : 60316 Mbit/s > > Current 4.11.0-rc6: 47491 Mbit/s > > This patch : 60662 Mbit/s > (should instead state "Jesper's patch" or "His patch") > Yes, you are correct of course. > Ran same test (8 parallel netperf TCP_STREAMs) with this patch applied: > > This patch 60106 Mbit/s (average of 7 iteration 60 sec runs) > > With these speeds I'm starting to hit the memory bandwidth of my machines. > Thus, the 60 GBit/s measurement cannot be used to validate the > performance impact of reverting this compared to my softirq patch, it > only shows we fixed the regression. (I'm suspicious as I see a higher > contention on the page allocator lock (4% vs 1.3%) with this patch and > still same performance... but lets worry about that outside the rc-series). > Well, in itself that limitation highlights that evaluating this is challenging and needs careful treatment. Otherwise two different approaches can seem equivalent only because a hardware-related bottleneck was at play. -- Mel Gorman SUSE Labs -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
