docs: include required chunk size in perf readme #324
Conversation
I almost changed the default to something bigger because it improves the benchmark results.. It's probably worth stating explicitly what the chunk size is so every implementation uses the same value for consistency.
There was a problem hiding this comment.
In what way do you expect the high level application chunk size to matter?
We do chunking at the TCP level, Noise level and Yamux level. We do chunking at the QUIC level. Once the application (here the perf protocol) provides a continuous flow of data, i.e. fill the pipe, why does it matter whether that is happening in 64KiB or 128KiB chunks?
Great. Then I suggest increasing it in js-libp2p. That said, worth investigating why js-libp2p is CPU bottlenecked on the chunk size? |
The point of this PR is so that all implementations use the same chunk size so there is a better basis for comparison. Is this not something we want? |
I almost changed the default to something bigger because it improves the benchmark results..
It's probably worth stating explicitly what the chunk size is so every implementation uses the same value for consistency.