buglets with documentation of units in results.json

[RobertHenry6bev]

The results.json file contains, after a fashion, documentation on all of the measured quantities. Phew!

The documentation is in places inconsistent or incomplete.

Printing of benchmarks/cpu/raw prints far more digits of precision than the underlying system is capable of measuring.

Some times are reported in ms, and other times in us, and rates in per-second. I'd pick either ms or us, and stick with per-second.

Pedantic: if you report in us, perhaps use UTF8 mu symbol rather than Latin u.

Many, but not all, longDescriptions have units. All should have units, replicating what's already in shortDescription. See benchmarks/build-time, benchmarks/start-time, benchmarks/published-size where the units are omitted.

Some shortDescriptions use '(ms)' but the longDescription says 'in ms'. Use the style '(...)' uniformly.

Are sizes (such as benchmarks/swap) reported in SI Mega (1e6) or Mibi (1024*1024) units. This may be tricky.

The "format" is always given as "n0". I had to look n0 format up. That's irrelevant, since this is json, and json doesn't support using ',' to group digits. I don't see the point in reporting "format".

What happens if the number being reported is mis-computed upstream as Inf or NaN. What happens when it is printed in the json file? (JSON doesn't support Inf or NaN...)

[sebastienros]

The results.json file contains, after a fashion, documentation on all of the measured quantities. Phew!

You can exclude these from the results file by using --no-metadata and --no-measurements respectively

The documentation is in places inconsistent or incomplete.

Thanks for your recent help on that. Please file specific issues when you find some.

Printing of benchmarks/cpu/raw prints far more digits of precision than the underlying system is capable of measuring.

#100

Some times are reported in ms, and other times in us, and rates in per-second. I'd pick either ms or us, and stick with per-second.

I am following what dotnet counters export. And some clients (bombardier, wrk) use varying units too.
But filed this issue to follow up, we might be able to decide what common unit to use #101

Pedantic: if you report in us, perhaps use UTF8 mu symbol rather than Latin u.

Noted

Many, but not all, longDescriptions have units. All should have units, replicating what's already in shortDescription. See benchmarks/build-time, benchmarks/start-time, benchmarks/published-size where the units are omitted.
Some shortDescriptions use '(ms)' but the longDescription says 'in ms'. Use the style '(...)' uniformly.

Created #102

I just copied the dotnet counters descriptions. Could be better

Are sizes (such as benchmarks/swap) reported in SI Mega (1e6) or Mibi (1024*1024) units. This may be tricky.

Noted

The "format" is always given as "n0". I had to look n0 format up. That's irrelevant, since this is json, and json doesn't support using ',' to group digits. I don't see the point in reporting "format".

This format is not for description purpose, but for output formatting on the controller. It's just UI metadata.
Note that right now it also has data with n02. And json is also used for Bombardier's raw results.

What happens if the number being reported is mis-computed upstream as Inf or NaN. What happens when it is printed in the json file? (JSON doesn't support Inf or NaN...)

It should be stored as a string in this case.