Less Slow

This week in rngo 0.37.0, I made rngo less slow.

#Timing

I updated rngo to record timing metadata, specifically the simulation start and end times. So, I was able to add that to the audit for my semi-realistic application Takeoff. It was slow:

Simulation
time: 2026-09-26 13:41:58
db: 31317 effects, 0 outputs

Audit
elapsed-seconds: 564.551

So about 10 minutes for 31k inputs, which is very slow. I set up some Criterion benchmarks and addressed some low-hanging fruit and got it down to 10 seconds:

Simulation
time: 2026-09-27 02:37:43
db: 31317 effects, 0 outputs

Audit
elapsed-seconds: 10.37

The main improvements were adding indexes to the SQLite database, but avoiding database calls by caching data in an in-process data store.

The obvious concern with caching is how it will impact the process's memory. I used a Fenwick tree implementation, along with some other optimizations, which means that 100 million events will require less than 10 MBs of memory, which I think is acceptable for now.

#Library Structure

I also improved the organization of the rngo Rust library, first of all by getting rid of the rngo-sim crate and putting all the code in the top-level rngo crate. I also consolidated a lot of code under the main effect, proxy, log and audit modules.

There will probably be more breaking changes coming, but all in the name of ultimately stabilizing on a high-quality, idiomatic library interface.

#Looking Forward

Next week, I plan on continuing to improve simulation realism and developer experience in the example Takeoff application.