🚦

Mining with less GPU but more PCD

Last edited
Feb 10, 2022 4:45 PM
Project
Storage on-boarding

Authors: @Kubuxu, @Irene , @Nicola

Main Problem: Sealing is expensive (in terms of execution, since it requires expensive hardware)

Goals

  • Decrease by half the GPU utilization

Intuition: Let’s decrease the cost of sealing

  • Attempt 1: Reduce the SNARK load by removing SHA
    • too difficult, requires trusted setup, requires new PoRep
  • Attempt 2: Off-chain PoRep verification with external network of challenger nodes
    • too difficult, but a good idea
  • Attempt 3: Explore new ways to mine without luxurious hardware (utilize lower-end hardware in a better way)
    • feasible, but seems too difficult
  • Attempt 3: Let someone else seal for you (will decrease the upfront cost, but won’t reduce the cost)
  • Attempt 4: Decrease the security of PoRep (even more!) to decrease the number of challenge, to decrese GPU overhead
    • feasible, very simple to implement

High level idea:

  • Decrease the PoRep security by decreasing the number of challenges, which decreases the number of SNARKs
  • In order to guarantee security in the network we must increase PCD
  • Having a higher PCD is an ok outcome

Pros & Cons

  • Pros:
    • ~90% of PoRep is SNARK generation, reducing the number of SNARKs will have a real impact.
  • Cons:
    • However, Snark technologies has improved of 10x during the past few years, if we believe the trend continues, then this solution might be not that useful

Formulas:

  • How many challenges (hence how many SNARK proofs) do we need for bits of security? (reducing s)

: bits of security (today: 10)

: fraction of errors (today: 0.039)

  • How much spacegap are we willing to give up? (by increasing x)