benchmark: calibrate util.*

[brunocroh]

According to nodejs/performance#186

Update all benchmarks in the util namespace by setting the n value to the result from calibrate-n.

Util Calibration Benchmarks

Util Function	Duration (minutes)	From Value	To Value
util.text-encoder	42.00	1e6	1e3
util.deprecate	1.00	1e5	1e3
util.format	2.00	1e6	1e2
util.get-callsite	5.50	1e6	1e2
util.inspect-array	1.30	5e3	10
util.inspect-proxy	0.47	1e5	1e2
util.inspect	5.14	8e4	1e3
util.parse-env	0.34	3e4	10
util.priority-queue	0.04	1e5	100
util.splice-one	2.12	5e6	1e3
util.style-text	0.48	1e3	1e2
util.type-check	0.28	1e6	1e3

[nodejs-github-bot]

Review requested:

• @nodejs/performance

[brunocroh]

I see some benchmarks using syntax like 1e3 and others using 1000. Which one is preferable, or does it not matter?

[Uzlopak]

1e3 is shorter than 1000, thats all.

[RafaelGSS]

@brunocroh can you share the machine you used to run calibrate-n? We tend to use the more dedicated machine possible. I had to set up a hetzner 4vCPUs 16gb dedicated machine for my recent benchmark updates.

If your machine is at least twice as good as the one I mentioned, it might get a different CV% on these machines. If that's the case, I suggest you increase by one 0 for the ones that reduced drastically, e.g: 8e4 -> 1e3

[brunocroh]

@brunocroh can you share the machine you used to run calibrate-n? We tend to use the more dedicated machine possible. I had to set up a hetzner 4vCPUs 16gb dedicated machine for my recent benchmark updates.

If your machine is at least twice as good as the one I mentioned, it might get a different CV% on these machines. If that's the case, I suggest you increase by one 0 for the ones that reduced drastically, e.g: 8e4 -> 1e3

Nice, I ran it on a Mac M2 with 12 CPU cores and 16 GB of RAM.
I also have a VPS that I can use for that purpose, 4 vCPUs and 12 GB of RAM. Let me retest it there.

[brunocroh]

@brunocroh can you share the machine you used to run calibrate-n? We tend to use the more dedicated machine possible. I had to set up a hetzner 4vCPUs 16gb dedicated machine for my recent benchmark updates.

If your machine is at least twice as good as the one I mentioned, it might get a different CV% on these machines. If that's the case, I suggest you increase by one 0 for the ones that reduced drastically, e.g: 8e4 -> 1e3

I set up a 4 vCPU, 16 GB dedicated machine to test these benchmarks again, and I reran the ones that took more than 5 minutes (text-encoder, inspect and get-callsite) to see if we could reduce n somehow, but unfortunately I couldn’t find any window for it.

So I think the best option is to close this PR without any changes for now.