MARF Caching Strategy without using RAM

[gregorycoppola]

Background

In the stacks-hyperchains repo, we are going to implement a MARF strategy that is optimized for the "no fork case" (hirosystems/stacks-subnets#6).

We are planning to use a sqlite-backed database that is:

1. stored in DB (not RAM)
2. not able to handle forks

Caching Strategy

Aside from or in addition to caching more of the MARF in RAM (#3014, #3088), we can try using this new database-based solution as a caching level.

Strategy:

1. Try hitting the non-MARF DB. If the existing state is valid for the current block, use this value.
2. Otherwise, do a full MARF look up.

The viability of this strategy would depend on the empirical distribution of:

• reads
• writes
• over-writes
• forks

The collection of data might suggests speed ups from this strategy.