Re: [PATCH v2][RFC] sched/fair: Change SIS_PROP to search idle CPU based on sum of util_avg
From: Chen Yu
Date: Sat Apr 09 2022 - 11:16:02 EST
On Tue, Apr 5, 2022 at 11:20 AM K Prateek Nayak <kprateek.nayak@xxxxxxx> wrote:
>
> Hello Chenyu,
>
> This is the output of vmstat captured every minute during the runtime of the benchmark:
>
> procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
> r b swpd free buff cache si so bi bo in cs us sy id wa st
> 0 0 0 524179712 47956 623328 0 0 18 1 95 70 0 1 99 0 0
> 0 0 0 524583488 6984 296760 0 0 869 32 765 613 0 0 100 0 0
> 2 0 0 519057408 10840 413092 0 0 1413 97 40022 245964 2 1 98 0 0
> 13 0 0 515806048 11036 1560292 0 0 1 15861 47291 481310 3 1 95 0 0
> 0 0 0 519494240 11844 2073552 0 0 27 11934 7397 70297 1 0 99 0 0
> 49 1 0 508753216 12064 4519468 0 0 22 82723 160775 1219361 11 5 84 0 0
> 51 1 0 508787424 12188 4686136 0 0 3 177661 199218 1522425 13 5 81 0 0
> 51 0 0 508717088 12464 5102900 0 0 8 172393 215555 1526566 14 6 81 0 0
> 54 0 0 508974208 12584 5133116 0 0 0 196022 194294 1520359 14 5 81 0 0
> 49 0 0 509171712 12720 5133420 0 0 1 189473 181195 1520990 14 5 81 0 0
> 65 2 0 509624672 12820 5172400 0 0 0 192406 190311 1516981 14 5 80 0 0
> 50 1 0 510052864 14228 5231312 0 0 886 173806 236733 1491140 14 6 80 0 0
> 50 1 0 510455904 14340 5231412 0 0 0 191996 252078 1479435 14 6 80 0 0
> 51 0 0 510845184 14464 5231436 0 0 0 191033 251433 1480865 14 6 80 0 0
> 54 0 0 511137472 15128 5247844 0 0 12 198460 254401 1485615 14 6 80 0 0
> 54 0 0 511242816 15240 5247868 0 0 0 205262 255299 1483994 14 6 80 0 0
> 0 0 0 524682656 6488 302624 0 0 596 64012 55117 315385 3 1 96 0 0
> 11 0 0 523307136 10096 350728 0 0 788 89 3350 30092 0 0 100 0 0
> 9 0 0 516713440 10276 1222368 0 0 2 8941 51751 497130 3 1 95 0 0
> 0 0 0 519524608 10428 2009192 0 0 8 18969 29798 277747 2 1 97 0 0
> 50 0 0 510469184 10608 4463716 0 0 15 72376 99893 592582 6 2 92 0 0
> 50 1 0 510225728 10760 4954988 0 0 3 175225 223641 1531010 13 5 81 0 0
> 49 0 0 510185376 10856 4955084 0 0 0 167856 208568 1541377 13 5 81 0 0
> 51 1 0 510101696 10964 5058408 0 0 0 185400 216839 1534992 13 5 81 0 0
> 56 0 0 509908448 11084 5058524 0 0 0 184871 240666 1478386 14 6 81 0 0
> 57 0 0 509593088 11188 5058624 0 0 0 174319 257078 1472289 14 6 80 0 0
> 46 0 0 509449280 11320 5120364 0 0 0 199695 246095 1484076 14 6 81 0 0
> 47 0 0 509569120 11440 5120504 0 0 0 195409 246954 1486632 14 6 81 0 0
> 51 1 0 509394240 11564 5120640 0 0 0 187739 235864 1487453 14 5 81 0 0
> 46 1 0 509165184 11668 5137480 0 0 0 190097 225997 1510812 14 5 81 0 0
> 83 1 0 509214944 11788 5137564 0 0 0 185712 232513 1524289 14 5 81 0 0
> 0 0 0 524685792 3940 268116 0 0 48 153174 152934 966090 9 3 88 0 0
> 0 0 0 524574880 6560 317068 0 0 703 82 193 342 0 0 100 0 0
> 11 0 0 515208960 8380 877736 0 0 606 8627 31043 324947 2 1 97 0 0
> 1 0 0 519673440 9196 1845944 0 0 20 13871 41874 470542 3 1 96 0 0
> 56 0 0 513421280 9328 2819588 0 0 13 6776 21005 106752 1 0 98 0 0
> 46 0 0 509829408 9520 4192212 0 0 4 157023 199117 1512516 13 5 82 0 0
> 49 1 0 509403200 10084 4596844 0 0 22 175970 205070 1512552 13 5 81 0 0
> 50 0 0 508933536 10220 4828368 0 0 0 180676 224455 1512344 14 6 81 0 0
> 47 0 0 508529472 10364 4940972 0 0 1 196158 250799 1502868 14 6 80 0 0
> 58 0 0 508565344 10476 4941012 0 0 0 194334 256721 1502510 14 6 80 0 0
> 50 0 0 508594176 10596 4941144 0 0 0 182299 257354 1501533 14 6 80 0 0
> 57 1 0 508721088 10700 4965532 0 0 0 186060 255378 1501167 14 6 80 0 0
> 49 0 0 508646144 10824 4965644 0 0 0 191029 257866 1500889 14 6 80 0 0
> 49 0 0 508739136 10920 4965672 0 0 0 182014 260923 1504201 14 6 80 0 0
> 50 0 0 509059616 11032 4974112 0 0 0 195981 260905 1503577 14 6 80 0 0
> 0 1 0 516111104 11152 4975472 0 0 0 190756 186797 1452580 13 5 82 0 0
> 0 0 0 519977088 17876 4985004 0 0 177 18948 747 650 0 0 100 0 0
>
> If there is any specific monitoring program you would like me to run, please let me know.
>
Thanks for providing this information. So the ycsb-mongodb had some
regression when the
load was around 20%. As v2 patch would change the search depth when
the load is low, this
results might indicate that we should search more CPUs when the system is low.
thanks,
Chenyu