Support starting benchmark tasks using taskset on Linux

[mcollina]

Our benchmark tasks must be modified to add a taskset command to pin the CPU the benchmark will run on. This is needed because now CPUs come with performance cores and energy efficient cores, making it harder to have reliable results otherwise.

It's also needed by nodejs/build#3657.

node/benchmark/compare.js

Lines 73 to 75 in af48641

	const child = fork(path.resolve(__dirname, job.filename), cli.optional.set, {
	execPath: cli.optional[job.binary],
	});

https://github.com/nodejs/node/blob/main/benchmark/run.js#L43-L46

cc @nodejs/performance

[ryanaslett]

FYI: It appears as though the p cores are 2 threads each, and the e cores are only one. This is the output of lscpu on one of the machinees:

root@test-hetzner-ubuntu2204-x64-2 ~ #  lscpu --all --extended
CPU NODE SOCKET CORE L1d:L1i:L2:L3 ONLINE    MAXMHZ   MINMHZ      MHZ
  0    0      0    0 0:0:0:0          yes 4800.0000 800.0000 2500.000
  1    0      0    0 0:0:0:0          yes 4800.0000 800.0000 2500.000
  2    0      0    1 4:4:1:0          yes 4800.0000 800.0000 2500.000
  3    0      0    1 4:4:1:0          yes 4800.0000 800.0000 2500.000
  4    0      0    2 8:8:2:0          yes 4800.0000 800.0000 2500.000
  5    0      0    2 8:8:2:0          yes 4800.0000 800.0000 2500.000
  6    0      0    3 12:12:3:0        yes 4800.0000 800.0000 2500.000
  7    0      0    3 12:12:3:0        yes 4800.0000 800.0000 2500.000
  8    0      0    4 16:16:4:0        yes 4800.0000 800.0000 2500.000
  9    0      0    4 16:16:4:0        yes 4800.0000 800.0000 2500.000
 10    0      0    5 20:20:5:0        yes 4800.0000 800.0000  800.047
 11    0      0    5 20:20:5:0        yes 4800.0000 800.0000 2500.000
 12    0      0    6 24:24:6:0        yes 3500.0000 800.0000 2500.000
 13    0      0    7 25:25:6:0        yes 3500.0000 800.0000 2500.000
 14    0      0    8 26:26:6:0        yes 3500.0000 800.0000 2500.000
 15    0      0    9 27:27:6:0        yes 3500.0000 800.0000 2500.000
 16    0      0   10 28:28:7:0        yes 3500.0000 800.0000 2500.000
 17    0      0   11 29:29:7:0        yes 3500.0000 800.0000 2500.000
 18    0      0   12 30:30:7:0        yes 3500.0000 800.0000 2500.000
 19    0      0   13 31:31:7:0        yes 3500.0000 800.0000 2500.000

So it should just be a matter of taskset -c 0-11 <command> for the perf cores.

[RafaelGSS]

FWIW A long time ago I did research to see the impact of CPU 0 on Node.js tasks and apparently, it doesn't affect Node.js workload. https://github.com/RafaelGSS/lies-and-benchmark/blob/76a261381ab1fae378bc24ff1853d39e48e8bcde/cpu-0-variation/cpu-http.js#L25

I did a more comprehensive benchmark during that research, but I couldn't find any reference to it in my personal notes.

[thisalihassan]

I will make a PR for this

[mcollina]

@RafaelGSS the reason why this is needed is to remove variance from the tests, because on new machines there are different types of cores.