Content
@
0 reply
1 recast
2 reactions
The Mana Plug
@manaplug
Okay. I've been in a DEEP rabbit hole. I'll write it here so y'all can go tough grass and I'll go back to my cave. 3 days ago, core Solana devs have talked about how they know exactly what the problem is. And they took time off of firedancer development to work on it.
1 reply
0 recast
0 reaction
The Mana Plug
@manaplug
It's a packet prioritization and hardening issue with the "Agave" servers. Which collect user transactions and send them to validator to execute. These servers are attempting to process 100k to 10M packets per second. Even 1 microsecond on a packet is too long and each one needs cryptographic hashing ran on them.
1 reply
0 recast
0 reaction
The Mana Plug
@manaplug
As a dev just because you know what the problem is in a complex system, doesn't mean there is a simple fix. This is a gigantic production environment, and while congestion is a problem now, you would rather have degraded performance, than ZERO performance
1 reply
0 recast
0 reaction
The Mana Plug
@manaplug
They mentioned that this specific server uses QUIC as a network protocol and people have been quick to blame this, however they note that any protocol would be having these issues, specifically ones like this which obfuscate the data inside the packets, routers can't sniff them.
1 reply
0 recast
0 reaction
The Mana Plug
@manaplug
Currently there are so many trash transactions getting to these servers and there are no optimizations to dump garbage transactions because they were probably focused on network uptime and executing all transactions quickly.
1 reply
0 recast
0 reaction
The Mana Plug
@manaplug
However they have a conceptual plan moving forward. However it still needs to be executed. And they are doing it in an incredibly complex live environment.
1 reply
0 recast
0 reaction
