Created: 26th February 2023 by @Anonymous

This document is about how to run more benchmarks that may give more precise insights about the use of Lookups for Filecoin.

TL;DR: Somebody with engineering expertise may have to run them; this doc explains what/how

Background

caulk-dev-main(1).zip

The experiments to be run

The Sealing step in Filecoin consists of approximately 77K SHA computations (I will refer to this number as N). We will ignore non-sha computations. We will consider N’ = N/10 for Groth16 since what we do is to run 10 Groth16 in parallel in Filecoin.

The way to approximate this is is to run on the same machine:

How to draw a comparison and understand if this is a dead-end

If $T’ < T/4$ or less then it is possible to Lookup approach is promising (Baloo is unoptimized). Otherwise it is not worth it until lookup techniques improve.