Danny Price
@endat
0 reply
0 recast
0 reaction
0 reply
0 recast
0 reaction
0 reply
0 recast
2 reactions
0 reply
0 recast
0 reaction
0 reply
0 recast
0 reaction
0 reply
0 recast
0 reaction
0 reply
0 recast
0 reaction
0 reply
0 recast
0 reaction
0 reply
0 recast
1 reaction
0 reply
0 recast
0 reaction
0 reply
0 recast
0 reaction
3 replies
1 recast
3 reactions
0 reply
0 recast
0 reaction
0 reply
0 recast
0 reaction
0 reply
0 recast
0 reaction
0 reply
0 recast
0 reaction
0 reply
1 recast
19 reactions
0 reply
0 recast
0 reaction
A fun math aside, on the idea of splitting a large zk proving workload between multiple provers.
Suppose you have N provers, and you have a proving workload that you split into N parts (so, one part per prover). You require provers to pre-register, but registration is open-access.
Suppose you have a constant fault rate (eg. 1/5 of registered provers fail). Provers expect to complete in one round (eg. 3s). If one prover fails, other provers have to come in and re-prove that load. How many rounds does it take for the entire workload to get proven?
Answer: log*(N)
(yes, that's the iterated-log function)
Why:
In the first round, you go from N unproven workloads to N/5 unproven workloads
In the second round, each remaining workload gets assigned 5 provers, so per-workload failure rate becomes 1 in 5^5. So you go to N / 5 / 5^5 unproven workloads
In the third round, each remaining workload gets assigned ~5^5 provers, so failure rate is 1 in 5^(5^5). So you go to N / 5 / 5^5 / 5^(5^5) unproven workloads 20 replies
27 recasts
246 reactions
13 replies
21 recasts
89 reactions
