Bench issues

[Hennzau]

Describe the bug
I wanted to bench dora again, but it seems a bit brocken now and I don't know exactly why.

To Reproduce
Steps to reproduce the behavior:

1. clone dora repository
2. cd dora
3. cargo run --example benchmark --release

Expected behavior
A good bench.

Behavior
Definitely not a good bench, and really strange.

Latency:
size 0x0     : 990.934µs
size 0x8     : 1.005274ms
size 0x40    : 1.031393ms
size 0x200   : 1.036277ms
size 0x800   : 1.025507ms
size 0x1000  : 986.606µs
size 0x4000  : 928.782µs
size 0xa000  : 950.716µs
size 0x64000 : 958.542µs
size 0x3e8000: 977.199µs
Throughput:
size 0x40    : 194 messages per second
size 0x800   : 614 messages per second
size 0x1000  : 57 messages per second
size 0x4000  : 35 messages per second
size 0x64000 : 1503 messages per second
size 0x3e8000: 198 messages per second

Environments (please complete the following information):

• System info: [use uname --all on Linux]

Linux enzo 6.8.0-51-generic #52-Ubuntu SMP PREEMPT_DYNAMIC Thu Dec  5 13:09:44 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux

• Dora version: [use dora --version and pip show dora-rs]

Latest main

[haixuanTao]

Ok, interesting, it might be related to the scheduler. I think one of our current bottleneck on speed is the fact that we need to pull data from the daemon using a request from the node.

I don't really have the bandwith to work on this but removing this should reduce latency.

[Hennzau]

Alright, yes it's not urgent!

[haixuanTao]

Fixed as of #971

We can open an additional one if we want to keep improving dora performance