[Question] is there a benchmark/comparison test to the standard libmysqlclient available?

[LowLevelMahn]

not the speed of the database itself but the overhead of protocol-handling etc. in speed and memory usage?

[anarthal]

Hi!

Not right now, although I'd like to add one. Are you aware of any standard tests I could perform? Or otherwise any specific comparison you'd like to see?

The points I don't see clear (and why I haven't done it yet):

• Should we use text queries or prepared statements? The extra round-trip of statement preparation/execution, which you can easily avoid using with_params.
• Should we include session establishment overhead somehow? Part of this overhead can be avoided withconnection_pool.
• Finally, should we use sync functions, async functions using callbacks, or async functions using coroutines? I'd inclide towards the last one.

I'm happy to hear your suggestions.

Thanks,
Ruben.

[LowLevelMahn]

Are you aware of any standard tests I could perform? Or otherwise any specific comparison you'd like to see?

any standard test i found are db-performance related - not the performance of the protocol encoding/decoding
in principle is it encoding-speed: large queries (with many parameter with large values , strings in select etc.) for prepared statements
and queries with large (many rows/cols, datasize) resultset (decoding-speed)

(maybe only using temporary tables - or what is best to keep the db performance out of the test)
Test 1: with libmysqlclient -> this is the Oracle-Speed
Test 2: the same queries using Boost.Mysql

additionally on procotol-level (no connection to the DB, just the buffers with binary data grabbed from Test 2 as input or regression test) just to check how fast the encoding/decoding behaves (if that is relevant at all)

Should we use text queries or prepared statements?

both but i think the prepared statements are more encoding speed relevant

the additional tests could use the Test1/2 queries

the problem is to keep the DB mostly out of the test - so maybe a temporary/virtual table or something were the speed is more constant