Steven Allen
Steven Allen
Yep, that was my conclusion as well.
1. Having the event subscription _only_ query the database makes total sense to me. I.e., process chain events once, put them in the database, then trigger queries over the _database_...
> During the triage we discussed if we could drop the cache (maybe once a day). Specifically, drop the state cache for all tipsets not on the canonical chain at...
Really, we can just: 1. Not return the cached value. 2. Set `recompute = true`. 3. Finish out the rest of the function.
(going to write a benchmark now to confirm that this is fixed)
Benchmark in https://github.com/filecoin-project/ref-fvm/pull/2060, but I'm not seeing any compilation time changes between this version, v4.4, and v4.3.
Measurement strategy.... 1. Perform setup. 2. Log nothing 10 times. 3. Perform the test (ideally _many_ times). 4. Log once. We can use the first 10 logs to determine accurate...
For now, I'm going to _assume_: 1. 0.5gas per memory byte copied and/or filled. 2. 100gas per random read (10ns which is about the random access speed of most DDR4...
Look into the affects of non-aligned memcpy and aligned memcpy.
We've benchmarked this in the EVM (#1281) and have determined that it won't be an issue in M2.1. So we're punting to M2.2