Can you add this code to your test to

[tender hound]

I've got an entire subsystem dedicated to profiling timing and gc usage, so if you want the firehose it's available. I've also got over a thousand lined of benchmarking code primarily focused on identifying the specific use cases which cause temporary allocations the GC will need to deal with.

But the short answer is that an example project I've recently been working on, after everything is fully loaded and the full HTTP/Rest/WebSockets server has spun up uses roughly half (i.e. 1MB) of the ram on on an S2 Mini (which has 2MB ram / 4MB flash)

Once it's in "steady state" - with an active WebUI page open and updating over WebSockets - It then settles into a cycle of running a little over a minute before hitting the low threshold (around 600K free) and triggering a gc.collect(), which takes it back to around 700K used (1300 free).