Messages in this thread | | | Date | Sun, 29 Jun 2014 15:47:00 -0700 | From | "Paul E. McKenney" <> | Subject | Re: [rcu] 2d033d5c0d4: +225.2% iperf.tcp.sender.bps |
| |
On Sun, Jun 29, 2014 at 11:15:44PM +0800, Fengguang Wu wrote: > Hi Paul, > > FYI, we noticed the below changes on > > git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git rcu/next > commit 2d033d5c0d424b7029abd0fc82e940ebc318fd89 ("rcu: Bind grace-period kthreads to non-NO_HZ_FULL CPUs")
Nice! Clearly outlines the hazards of providing too few housekeeping CPUs, I would guess.
Thanx, Paul
> test case: bens/iperf/300s-tcp > > e17adc3f20bc9a1 2d033d5c0d424b7029abd0fc8 > --------------- ------------------------- > 7.011e+09 ~ 1% +225.2% 2.28e+10 ~ 2% TOTAL iperf.tcp.sender.bps > 7.011e+09 ~ 1% +225.2% 2.28e+10 ~ 2% TOTAL iperf.tcp.receiver.bps > 15620691 ~ 1% +224.3% 50663012 ~ 2% TOTAL proc-vmstat.pgalloc_normal > 62969722 ~ 1% +224.3% 2.042e+08 ~ 2% TOTAL proc-vmstat.pgfree > 47347349 ~ 1% +224.3% 1.535e+08 ~ 2% TOTAL proc-vmstat.pgalloc_dma32 > 4996590 ~ 1% +218.9% 15933563 ~ 3% TOTAL softirqs.NET_RX > 8084072 ~ 1% +218.4% 25739055 ~ 2% TOTAL proc-vmstat.numa_hit > 8084072 ~ 1% +218.4% 25739055 ~ 2% TOTAL proc-vmstat.numa_local > 28676 ~ 2% -47.5% 15059 ~11% TOTAL softirqs.RCU > 99756 ~ 2% +76.0% 175606 ~ 2% TOTAL softirqs.SCHED > 562 ~11% +50.9% 848 ~17% TOTAL slabinfo.proc_inode_cache.active_objs > 620 ~ 8% +38.7% 860 ~15% TOTAL slabinfo.proc_inode_cache.num_objs > 271905 ~ 4% +17.4% 319216 ~ 0% TOTAL softirqs.TIMER > 1117 ~ 1% -9.1% 1015 ~ 2% TOTAL proc-vmstat.pgactivate > 9049 ~11% -94.6% 485 ~26% TOTAL time.involuntary_context_switches > 23233 ~ 1% +72.1% 39979 ~ 2% TOTAL vmstat.system.cs > 13078 ~ 1% +69.1% 22117 ~ 1% TOTAL vmstat.system.in > > Legend: > ~XX% - stddev percent > [+-]XX% - change percent > > > iperf.tcp.sender.bps > > 2.4e+10 ++----O-----O-----------O--------------O-----O--------O-----O-----+ > | O O O O O O O O | > 2.2e+10 O+ O O O | > 2e+10 ++ O | > | | > 1.8e+10 ++ | > 1.6e+10 ++ O O | > | | > 1.4e+10 ++ | > 1.2e+10 ++ | > | | > 1e+10 ++ | > 8e+09 ++ | > *..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..* > 6e+09 ++----------------------------------------------------------------+ > > > iperf.tcp.receiver.bps > > 2.4e+10 ++----O-----O-----------O--------------O-----O--------O-----O-----+ > | O O O O O O O O | > 2.2e+10 O+ O O O | > 2e+10 ++ O | > | | > 1.8e+10 ++ | > 1.6e+10 ++ O O | > | | > 1.4e+10 ++ | > 1.2e+10 ++ | > | | > 1e+10 ++ | > 8e+09 ++ | > *..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..*..* > 6e+09 ++----------------------------------------------------------------+ > > > time.involuntary_context_switches > > 14000 ++------------------------------------------------------------------+ > | | > 12000 *+. .*.. ..*..*.. .*.. | > | .*..*. *. *. | > 10000 ++ *. *..*.. .*..*..*.. ..*..*.. * > | *. *. .*.. ..| > 8000 ++ *. * | > | | > 6000 ++ | > | | > 4000 ++ | > | | > 2000 ++ | > | O O O O | > 0 O+-O--O--O--O--O---O--O--O--O--O--O--O--O-----O------O--O-----O-----+ > > > [*] bisect-good sample > [O] bisect-bad sample > > > Disclaimer: > Results have been estimated based on internal Intel analysis and are provided > for informational purposes only. Any difference in system hardware or software > design or configuration may affect actual performance. > > Thanks, > Fengguang
> ./iperf3 -s > ./iperf3 -t 300 -f M -J -c 127.0.0.1
| |