Ben Coomes

# Next Steps As for what to do next - * Would you welcome a PR for caching marshaled bytes? I can open one for that. * Would you welcome...

> Load testing with a "normal" machine, e.g. one with a common proportion between CPU and Memory, even with the estimator off by a factor of 2x shows that SpiceDB...

Hey @jakedt! I ran them Friday but haven't analyzed the results yet. I'll post a summary here later today, and I think I'll be able to give more details in...

# Summary Looking over the load tests, I don’t think that either the ‘accurate cost function’ or ‘marshaling cache’ changes make sense to implement. However, it still may be good...

That's a good idea! I'll give that a try and see how it goes. But still, while the accurate costers controlled memory use correctly for a single local instance, they...

Calculating the cost and inserting on a goroutine helped a bit, but read latency was still 47% slower compared to baseline. I did confirm that the buffered channel was big...

Hi @dblock I'd like to take a shot at this, as I need to learn Ruby and am familiar with the Strava API. However, I can't think of a good...

Okay, that makes sense. Just thinking about what each model class represents, `activity_total`, `activity`, `explorer_segment`, `lap`, `route`, `segment`, and `split` could all have a `total_elevation_loss` property. And, all except `explorer_segement`...