Messages in this thread | | | Subject | Re: [PATCH] watchdog: nohz: don't run watchdog on nohz_full cores | From | Mike Galbraith <> | Date | Tue, 31 Mar 2015 08:34:17 +0200 |
| |
On Tue, 2015-03-31 at 04:04 +0200, Mike Galbraith wrote:
> A bigger deal is the clocksource watchdog methinks. Measurement > inspired me to make it dead yesterday.
This is why btw. Note the alternating sum/s for 1 second samples.. shooting the watchdog makes that a steady 5us.
homer:~ # cgexec -g cpuset:rtcpus taskset -c 3 pert 1 3591.70 MHZ CPU perturbation threshold 0.013 usecs. pert/s: 4 >5.61us: 2 min: 0.04 max: 6.89 avg: 3.68 sum/s: 15us overhead: 0.00% pert/s: 3 >5.34us: 0 min: 0.04 max: 4.01 avg: 1.79 sum/s: 5us overhead: 0.00% pert/s: 5 >5.33us: 2 min: 0.04 max: 7.34 avg: 3.11 sum/s: 16us overhead: 0.00% pert/s: 3 >5.08us: 0 min: 0.04 max: 3.87 avg: 1.75 sum/s: 5us overhead: 0.00% pert/s: 3 >5.18us: 2 min: 1.04 max: 7.16 avg: 5.10 sum/s: 15us overhead: 0.00% pert/s: 4 >4.93us: 0 min: 0.04 max: 3.93 avg: 1.33 sum/s: 5us overhead: 0.00% pert/s: 6 >4.91us: 2 min: 0.04 max: 7.05 avg: 2.55 sum/s: 15us overhead: 0.00% pert/s: 3 >4.70us: 0 min: 0.04 max: 3.73 avg: 1.78 sum/s: 5us overhead: 0.00%
| |