On Tue, 2015-05-05 at 15:00 +0200, Thomas Gleixner wrote:
On Sat, 25 Apr 2015, Eric Dumazet wrote:
On Thu, 2015-04-23 at 14:45 +0200, Thomas Gleixner wrote:
You definitely have a point from the high throughput networking perspective.
Though in a power optimizing scenario with minimal network traffic this might be the wrong decision. We have to gather data from the power maniacs whether this matters or not. The FULL_NO_HZ camp might be pretty unhappy about the above.
Sure, I understand.
To make this clear, here the profile on a moderately loaded TCP server, pushing ~20Gbits of data. Most of TCP output is ACK clock driven (thus from softirq context).
(using regular sendmsg() system calls, that why the get_nohz_timer_target() is 'only' second in the profile, but add the find_next_bit() to it and this is very close being at first position)
PerfTop: 4712 irqs/sec kernel:96.7% exact: 0.0% [4000Hz cycles], (all, 72 CPUs)
10.16% [kernel] [k] copy_user_enhanced_fast_string 5.66% [kernel] [k] get_nohz_timer_target 5.59% [kernel] [k] _raw_spin_lock 2.53% [kernel] [k] __netif_receive_skb_core 2.27% [kernel] [k] find_next_bit 1.90% [kernel] [k] tcp_ackMaybe a reasonable heuristic would be to change /proc/sys/kernel/timer_migration default to 0 on hosts with more than 32 cpus.
profile with timer_migration = 0
PerfTop: 3656 irqs/sec kernel:94.3% exact: 0.0% [4000Hz cycles], (all, 72 CPUs)
13.95% [kernel] [k] copy_user_enhanced_fast_string 4.65% [kernel] [k] _raw_spin_lock 2.57% [kernel] [k] __netif_receive_skb_core 2.33% [kernel] [k] tcp_ackIs that with the static key patch applied?
This was without.
I applied your patch on current linus tree, but for some reason my 72 cpu host is not liking the resulting kernel. I had to ask for a repair, and this might take a while.
Note your kernel works correctly on other hosts, but with 48 or 32 cpus, so this must be something unrelated.
I'll let you know when I get more interesting data.
Thanks