Run Yjs crdt-benchmarks

composablesys / collabs

Collabs library monorepo

https://collabs.readthedocs.io/

Apache License 2.0

245 stars 11 forks source link

Run Yjs crdt-benchmarks #164

Open mweidner037 opened 2 years ago

mweidner037 commented 2 years ago

Besides the macro automerge_perf benchmark, https://github.com/dmonad/crdt-benchmarks contains text microbenchmarks. We should fork that repo and run it using our CText.

mweidner037 commented 1 year ago

In fact we should do the full suite and make a PR to that repo.

mweidner037 commented 1 year ago

Here: https://github.com/mweidner037/dmonad-crdt-benchmarks/ I used CRichText instead of CText for a fairer comparison with Y.Text.

The receiveTime branch takes a few extra measurements.

Results are good except:

Very long time for prepend microbenchmark (B1.3).
B4: ~8x slower sends than Yjs (especially noticeable on B4x100), but ~4x faster receives which is what matters in practice.

I'll wait to make a PR to the original repo until we reach Collabs 1.0.