add insert benchmarks comparing myxql/postgrex/exqlite

[kevinlang]

These benchmarks are forked from ecto_sql.

I only ran the Insert benchmarks, so will need to update the other benchmarks at some point and run them too. Was interested in Insert since SQLite can only have one active writer. I knew that SQLite in WAL mode was faster than most know, but the results nevertheless surprised me.

Results from running:

MIX_ENV=bench mix run bench/scripts/macro/insert_bench.exs 

##### With input Changeset #####
Name                     ips        average  deviation         median         99th %
Exqlite Insert       7759.47       0.129 ms    ±33.34%       0.118 ms        0.24 ms
Pg Insert             486.46        2.06 ms     ±9.79%        2.03 ms        2.61 ms
MyXQL Insert          266.38        3.75 ms    ±10.71%        3.72 ms        5.32 ms

Comparison: 
Exqlite Insert       7759.47
Pg Insert             486.46 - 15.95x slower
MyXQL Insert          266.38 - 29.13x slower

##### With input Struct #####
Name                     ips        average  deviation         median         99th %
Exqlite Insert       8710.79       0.115 ms    ±30.08%       0.104 ms        0.23 ms
Pg Insert             485.47        2.06 ms     ±9.74%        2.04 ms        2.64 ms
MyXQL Insert          264.55        3.78 ms    ±37.33%        3.70 ms        5.58 ms

Comparison: 
Exqlite Insert       8710.79
Pg Insert             485.47 - 17.94x slower
MyXQL Insert          264.55 - 32.93x slower

Wow!

[kevinlang]

Specs:

Operating System: Linux
CPU Information: AMD Ryzen 7 PRO 4750U with Radeon Graphics
Number of Available Cores: 16
Available memory: 14.92 GB
Elixir 1.11.3
Erlang 23.2.6

Disk is a 1TB Sabrent Rocket Q SSD.

[warmwaffles]

That is amazing.

Did you try it without WAL?

[kevinlang]

Nope, I'd be curious to see that too, just for comparison, even if I can't think of situations I wouldn't use WAL.

I hope to fix up the rest of the benchmark tests and run each test suite for comparison, probably once the rest of the main "1.0" functional issues are done. The "micro" ones they have around testing how long the adapter takes to build up the query is pretty interesting, too.

The main reason I wanted this driver was for use in moderate size applications, but was quite worried about the insert speed. It's good to know that this little database can truly handle 99% of load profiles :)

[warmwaffles]

Yea I'm excited for it.

Wondering if we should pull together a docker compose file to benchmark the library repeatedly.