Memory usage optimization / avoid sending unneeded data to parallel processing

[PragTob]

Should combat memory consumption issues caused by sending too much data to processes.

edit: To highlight the severeness of this:

• there is a benchmark I'm working on that I almost couldn't run before this as my 32GB of RAM would not be enough
  • benchmark is this one: https://github.com/PragTob/tco_elixir
• it also hung forever before calculating statistics and running the formatters
• In that benchmark I used :save to compare performance across elixir version, before this PR each individual save file was 200+MB. With this MR, it's 1MB (as it removes the data from formatters and :save is a formatter under the hood). The file size isn't the major thing, but it illustrates how much memory we save and how much time copying memory we avoid.

I wanna write up all of this in like 2 blog posts but as per usual my speed is abysmal 😁

[PragTob]

Mental notes:

1. We can be even more efficient for the most common case of just 1 formatter if we just don't do any parallel processing in that case, difference is small but exists and should be an easy fix:

Name                        ips        average  deviation         median         99th %
sequential_output        240.79        4.15 ms     ±5.70%        4.13 ms        4.34 ms
format & write           240.00        4.17 ms     ±3.71%        4.13 ms        4.82 ms
Formatter.output         228.23        4.38 ms    ±16.58%        4.07 ms        6.34 ms

Comparison: 
sequential_output        240.79
format & write           240.00 - 1.00x slower +0.0138 ms
Formatter.output         228.23 - 1.06x slower +0.23 ms

2. might want to provide "input_names" to use instead of the broken list :eyes:

3. adjust docs & Changelog

[PragTob]

:partying_face:

Name                        ips        average  deviation         median         99th %
Formatter.output         248.71        4.02 ms     ±4.58%        4.03 ms        4.59 ms
sequential_output        248.38        4.03 ms     ±3.16%        4.00 ms        4.53 ms
format & write           247.77        4.04 ms     ±6.76%        4.02 ms        4.35 ms

Comparison: 
Formatter.output         248.71
sequential_output        248.38 - 1.00x slower +0.00539 ms
format & write           247.77 - 1.00x slower +0.0153 ms

[PragTob]

I definitely should get exguard working again... the amount of force pushes due to minor credo things are embarrassing ._.

[PragTob]

:wave: Hey @hrzndhrn @NickNeck as you have some of the most Benchee plugins I'd appreciate it if you took a look here, mainly from the PoV on impact for plugins. If you want to review the PR as a whole of course I'm happy too :grin:

TLDR;

• sending a lot of data between processes VERY BAD
• formatters are losing access to Scenario.function, Scenario.input and Configuration.inputs will only retain the names, not the values

Thanks!

[NickNeck]

@PragTob thanks for the hint. Thanks for Benchee and all your work. I will take a look during the weekend.

[PragTob]

@NickNeck thanks for all your great OSS work and also specifically of creating Benchee plugins, it was designed to facillitate and allow this but seeing all the things people do warms my heart! 💚

[PragTob]

@NickNeck thanks a lot! :green_heart: