add mimalloc

[Shourya742]

This PR switches the binaries to use mimalloc instead of the default glibc allocator, which can exhibit higher fragmentation and contention under our workload.

[GitGab19]

Why do we need this change?

Do we have some data which demonstrates how this change is improving performances of our apps?

[Shourya742]

Why do we need this change?

Do we have some data which demonstrates how this change is improving performances of our apps?

Yes, it’s a memory allocator that’s better suited to our use case: long-running processes with frequent allocations, high concurrency, and a need for consistent latency. The default global allocator isn’t ideal here, as it tends to increase fragmentation and introduce global locking during allocations. I will run benchmarks and share the results here.

[Shourya742]

Global Allocator:

mimalloc:

Mimalloc does make us faster.

[GitGab19]

Should we close this @Shourya742 ?

Or at least putting it in draft?

[Shourya742]

I don't remember now, why we stalled this PR?

[GitGab19]

I don't remember exactly now, but it was not something we ever planned?

Maybe it makes sense to run again the analysis after the improvements on hotpath described here?

[Shourya742]

That might be the case, once the hotpath thingy is in. I will give the numbers another shot.