Open HalosGhost opened 2 months ago
I've not yet addressed Anders's suggestion. Hoping to test that later this week.
@anders94 I did some initial (read: not thorough) testing reducing the validation-thread-count in the coordinators/shards by 1, and see negligible difference (if anything slightly higher latency / lower throughput); but I've not tested at particularly large configurations.
Do you have time later today or tomorrow to hop on a call and I can walk through the preliminary results I have and we can see if it feels like I should run a few more tests (i.e., just to eliminate the possibility of noise, or at larger configurations to potentially amplify the effects)?
Yep, let's do it.
This simply rebases and updates #189.
I've done a bit of local testing, and this dramatically reduces the impact of enabling sentinel attestations.
@anders94 had a suggestion to prefer
n_threads
==std::thread::hardware_concurrency() - 1
; I'm going to run a few tests to see if one or the other has an obvious performance benefit, and then I believe this is likely ready-to-merge.