Messages in this thread | | | Subject | Re: [PATCH 10/10] sched/fair: Implement an EEVDF like policy | From | Mike Galbraith <> | Date | Thu, 09 Mar 2023 05:23:33 +0100 |
| |
On Wed, 2023-03-08 at 14:36 +0100, Mike Galbraith wrote: > > Remember this little bugger, allegedly distilled from a real > application control thread starvation issue? > > 6.3.0.g8ca09d5-master > homer:/root # time taskset -c 3 starve > expecting to receive 10000000 signals > > real 0m24.424s > user 0m4.468s > sys 0m18.957s > > 6.3.0.g8ca09d5-eevdf > homer:/root # time taskset -c 3 starve > expecting to receive 10000000 signals > zzzzzz > ^C
Ok, seems there must be a math booboo lurking.
virgin source, 100% hog vs tbench buddy pair, all pinned.
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ P COMMAND 5060 root 20 0 4420 680 680 R 96.01 0.004 0:41.40 3 cpuhog 5058 root 20 0 25500 1920 1792 S 2.326 0.012 0:01.05 3 tbench 5059 root 20 0 8796 896 768 R 1.661 0.006 0:00.78 3 tbench_srv
echo NO_PRESERVE_LAG > features
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ P COMMAND 5060 root 20 0 4420 680 680 R 99.33 0.004 1:28.24 3 cpuhog 5058 root 20 0 25500 1920 1792 R 0.333 0.012 0:01.75 3 tbench 5059 root 20 0 8796 896 768 S 0.333 0.006 0:01.30 3 tbench_srv
| |