patrickdugan
commented
2 years ago Lihki did this in ~June 2021 and it, combined with the tech debt overhang and the non-FTX mixed margin like approach, plus the periodic nature of it, lead us to adopt a new model in August 2021.
Optimizations can follow, but let's get an initial benchmark of the performance of our settlement algorithm.
of Addresses # of Trades per Address Time to parse state on settlement block
10,000 100 N milliseconds 100,000 10 100,000 100 10,000 1000 100,000 1000
This last one is 100M vertices instead of 1M or 10M and likely larger than most 4 hour settlement graphs, but let's just see what it takes and if there's any convexity to the scaling of time-to-compute vs. number of vertices.