> On Apr 5, 2021, at 09:07, Stuart Henderson <s...@spacehopper.org> wrote:
> 
> I've attached r620-E5_2630v2-2p6c2t.tgz, from Dell PE R620 with E5-2630v2.
> This is a machine which has "disabling user TSC (skew=XXX)" reported for
> some cores:
> 
> Nov  5 16:08:34 pweb /bsd: cpu11: disabling user TSC (skew=-107)
> Nov  5 16:08:34 pweb /bsd: cpu13: disabling user TSC (skew=-101)

This is probably NUMA latency when synchronizing between packages.

If we want to account for this we will need to do more complicated stuff.

> 2 SMT per core, 6 core per package, 2 packages.
> 
> s-c-p in the below are smt/core/package for each cpu, i.e.
> cpu0: smt 0, core 0, package 0
> cpu1: smt 0, core 0, package 1
> cpu2: smt 0, core 1, package 0
> cpu3: smt 0, core 1, package 1
> etc.
> 
> cpu  s-c-p   Min           Max        Median           Avg        Stddev
> 1   0-0-1   -65           498           157       152.071     28.014331
> 2   0-1-0   -22            13            -4         -4.91     7.8085128
> 3   0-1-1   -42           507           112        116.39     27.882671
> 4   0-2-0   -14             7            -2        -0.272     4.7497975
> 5   0-2-1     3           380           137       138.336     18.935249
> 6   0-3-0   -25            17            -5        -4.035     6.9143028
> 7   0-3-1  -136           299            86        87.687     22.889101
> 8   0-4-0   -24             4           -10         -8.81     5.0030921
> 9   0-4-1  -355           353            -6        -25.56     69.909611
> 10   0-5-0     7            41            27        26.838     7.2798869
> 11   0-5-1  -313           163            19         7.541     62.852118
> 12   1-0-0   -30            15            -3         -6.58     9.1155193
> 13   1-0-1  -331           439          17.5         0.494     67.932708
> 14   1-1-0   -30            12            -9        -9.583     6.0745379
> 15   1-1-1  -358           432           -15       -28.774     74.476155
> 16   1-2-0   -28            14            -9        -9.666     6.8339701
> 17   1-2-1  -339           459            -9       -31.638     77.072817
> 18   1-3-0   -29            19            -8        -7.896     7.3815765
> 19   1-3-1  -211           289          -101       -59.981     63.882655
> 20   1-4-0   -21            28            -1        -1.064     7.7156617
> 21   1-4-1  -338           417         -24.5       -31.543     77.017062
> 22   1-5-0    -1            39            17        16.359     5.3022857
> 23   1-5-1  -352           447            28         8.155     69.831278
> 
> Grouping them by package/smt:
> 
>                N     Min      Max   Median           Avg        Stddev
> cpu-p0-smt0  5000     -25       41       -3        1.7622     14.367997
> cpu-p0-smt1  6000     -30       39       -5    -3.0716667     11.631154
> cpu-p1-smt0  6000    -355      507      100     79.410833      79.21407
> cpu-p1-smt1  6000    -358      459        4    -23.881167     75.312631
> 
> CPU has ITSC, I do not have any "clock going backwards" with monotime.c
> with kern.timecounter.hardware=tsc.
> 
> <r620-E5_2630v2-2p6c2t.tgz>

Reply via email to