Content
@
0 reply
0 recast
0 reaction
Cassie Heart
@cassie
This week at /quilibrium we shipped - A massive cascade of changes to the bootstrap peer code (mostly on the p2p stack side) to unlock sustained volume with improved message delivery under 100k simultaneous peers (accidentally proved this over 200k peers because my deploy script had an oops, rip my infra budget): https://github.com/QuilibriumNetwork/ceremonyclient/tree/v2.0-bootstrap - Heavy refactoring of blatant (lazy) abuse of defer/panic/recover in go-multiaddr, go-libp2p, go-libp2p-kad-dht, and go-libp2p-blossomsub to rip out huge chunks of stacked 200-600ns penalties that were blocking our targets for routing - passed the connection management/routing success tests supporting an overload of 100M messages/sec - Upstreamed the libp2p specific changes to our separate fork of libp2p stack so other teams can share in the perf wins: https://github.com/QuilibriumNetwork/go-libp2p-stack
19 replies
8 recasts
67 reactions
kk
@king
cassie, please let us know how good 100M messages/sec is compared to the current stat we're up against?
1 reply
0 recast
2 reactions
Cassie Heart
@cassie
Gossipsub (the messaging framework we forked from, used by filecoin and eth beacon chain consensus, and also partially for Farcaster) can sustain in the order of thousands of messages a second when all peers behave and network size is also in the low thousands. Going beyond that for either variable (peers in the network, messages per second) leads to message loss that increases exponentially the higher those variables rise.
1 reply
0 recast
5 reactions
