[ci] Use xlarge resource class for long running browser tests.

[brendandahl]

Testing to see how well the browser tests parallelize in CI.

[brendandahl]

	test-browser-chrome-2gb	test-browser-chrome-wasm64-4gb	test-browser-firefox
No Parallel (medium - 1 runner)
	0:46:16	0:42:50	0:36:09
	0:46:46	0:45:12	0:40:50
	0:45:36	0:43:26	0:34:26
	0:46:44	0:33:16	0:35:59
	0:48:29	0:44:50	0:34:02
average	0:46:46	0:41:55	0:36:17
Parallel (medium 4 - 2 runners)
	0:29:26	0:28:15	0:22:45
	0:28:00	0:27:00	0:21:42
	0:28:58	0:31:12	0:21:41
	0:26:58	0:26:29	0:21:57
	0:27:18	0:26:16	0:20:33
average	0:28:08	0:27:50	0:21:44
speedup from baseline	1.66	1.51	1.67
Parallel (xlarge 16 - 8 runners)
	0:09:49	0:09:29	0:07:30
speedup from baseline	4.76	4.42	4.84

[sbc100]

Given that several other non-browser test suites take ~30min it seems like spending more money to make the browser tests quicker would not be wroth it unless we also were to speed up those other test suites too.

I uploaded #25194 so see if I could speed up the new slowest runner: test-wasm64-4gb

[juj]

For curiosity, I tested different parallelism values on my Linux box:

16: Total core time: 1778.549s. Wallclock time: 114.699s. Parallelization: 15.51x.

32: Total core time: 2059.722s. Wallclock time: 70.208s. Parallelization: 29.34x.
32: Total core time: 2042.970s. Wallclock time: 70.036s. Parallelization: 29.17x.

64: Total core time: 2532.726s. Wallclock time: 49.703s. Parallelization: 50.96x.
64: Total core time: 2506.703s. Wallclock time: 49.115s. Parallelization: 51.04x.
64: Total core time: 2568.892s. Wallclock time: 50.292s. Parallelization: 51.08x.
64: Total core time: 2489.469s. Wallclock time: 49.365s. Parallelization: 50.43x
64: Total core time: 2471.243s. Wallclock time: 48.887s. Parallelization: 50.55x.
64: Total core time: 2550.984s. Wallclock time: 49.412s. Parallelization: 51.63x.

80: Total core time: 2817.407s. Wallclock time: 47.546s. Parallelization: 59.26x.
80: Total core time: 2848.213s. Wallclock time: 47.651s. Parallelization: 59.77x.
80: Total core time: 2805.875s. Wallclock time: 47.016s. Parallelization: 59.68x.

96: Total core time: 3016.244s. Wallclock time: 46.486s. Parallelization: 64.89x.
96: Total core time: 2976.783s. Wallclock time: 45.791s. Parallelization: 65.01x.
96: Total core time: 2931.232s. Wallclock time: 44.712s. Parallelization: 65.56x.
96: Total core time: 3005.507s. Wallclock time: 46.153s. Parallelization: 65.12x.
96: Total core time: 2939.468s. Wallclock time: 47.395s. Parallelization: 62.02x.
96: Total core time: 3307.454s. Wallclock time: 52.337s. Parallelization: 63.20x.

128: Total core time: 4522.871s. Wallclock time: 56.776s. Parallelization: 79.66x.
128: Total core time: 3951.026s. Wallclock time: 47.712s. Parallelization: 82.81x.
128: Total core time: 3866.550s. Wallclock time: 52.671s. Parallelization: 73.41x.
128: Total core time: 3629.438s. Wallclock time: 50.599s. Parallelization: 71.73x.
128: Total core time: 3704.257s. Wallclock time: 54.433s. Parallelization: 68.05x.
128: Total core time: 3727.838s. Wallclock time: 53.200s. Parallelization: 70.07x.

160: Total core time: 4832.477s. Wallclock time: 59.705s. Parallelization: 80.94x.
160: <failing, many tests fail on webgl context not being obtained>
160: <failing, many tests fail on webgl context not being obtained>

The system has 256 logical hardware threads, so this experiment showed that the sweet spot is at maybe 64 parallel runners, as adding more is wasting electricity for diminishing returns.

[brendandahl]

That was on browser tests?

[juj]

That was on browser tests?

Err, yeah, exactly. Running browser test suite with your awesome parallelised patch. Also, using Firefox as the test browser.