Proposal: Optimizing Signature Participation for Performance

[anhthii]

Currently, our signing operation includes all available nodes in the signing process. However, in tss-lib, only t+1 nodes are required to generate a valid threshold signature.

To reduce latency and network roundtrips, we propose changing the implementation to only involve t+1 nodes in the signing process — even when more nodes are online. This selective participation will significantly boost performance without compromising correctness or fault tolerance.

[tuananhk13bk]

The tradeoff between performance and fault tolerance is a very classic problem in distributed systems.

• With 3 nodes participating, the system tolerates 1 fault per signing round
• But with only 2 nodes participating, the system tolerates 0 faults.
  -> Is it worth sacrificing robustness and resilience for only a slight performance gain? @anhthii

[anhthii]

make sense but I think we must check if the chosen nodes are online before sending the sign request to these nodes.

So lets say that in a cluster of 3 nodes with t = 1, require t+1 = 2 nodes to sign then out of node A , B, C. First we might choose A, B if B is down, we chose A,C

[anhthii]

In tss the system only requires t+1 nodes to sign, meaning it can tolerate up to n−(t+1) nodes being offline.

[tuananhk13bk]

What if the chosen node is online at phase 1 (check) then goes down at phase 2 (sign)?

[anhthii]

it might rarely happen but if that is the case then it just return error