Memory requirements/challenges - any solutions?

[yanshay]

I'm developing an application that connects over TLS to some server.
My normal code use PSRAM everywhere using alloc, so whatever alloc used is for esp-wifi and esp-mbedtls (which uses the esp-wifi alloc) is the only allocations going to the internal RAM.

I noticed that for each TLS session I open about 30KB are used (+/-). Is that expected? Any way to use less? I need to open two more connections, maybe more at the same time and I don't have where to get memory from.

I'm still on esp-hal 0.23.1.

With all changes I saw recently taking place (which I guess need the newer esp-wifi and esp-hal), is esp-mbedtls maybe now can use the PSRAM for it's allocations?

[yanshay]

@AnthonyGrondin @ivmarkov Did you happen to see this question and know to answer?

[bjoernQ]

According to this https://esp32.com/viewtopic.php?t=10612 mbedtls itself will neeed around 23k per session. #75 might reduce memory consumption per session to get closer to that

[yanshay]

Thanks

And is there now a way to get that memory into PSRAM using alloc?

Currently I moved everything in my app to PSRAM including embassy tasks so the only limiting factor is that I need many SSL sessions that needs the internal RAM (which I guess isn't unique with everything being TLs today).

[ivmarkov]

It might be possible and in fact #71 #75 might even help doing that, in that it is switching esp-mbedtls to use a custom calloc/free pair of functions for all memory. This - in turn - would allow us to provide allocations to esp-mbedtls from PSRAM rather than from regular memory. Given that in general the mbedtls code is unaware of threads and atomics, this might even be safe (I wonder how you moved everything to PSRAM - given that doing it blindly is very unsafe; also I don't thinj it is possible to move "embassy tasks" to PSRAM, as these by design live in .bss, and I don't think .bss-in-PSRAM is supported, or is it?)

[ivmarkov]

@AnthonyGrondin Not really sure anymore what is the status with #75... Are you still on it? (update: was referring to the wrong PR)

[yanshay]

With regards to embassy it's possible by storing task's state in alloc memory using TaskStorage with Box::leak.

As for the everything I refer to is that any significant data item I alloc instead of other options.

[AnthonyGrondin]

@AnthonyGrondin Not really sure anymore what is the status with #75... Are you still on it? (update: was referring to the wrong PR)

I'll still thinking about #75 (comment) and how to better manage the state. I did update my app with the current implementation to test it and it's much better. That being said, I think in the end we should pass certs by reference to Certificates or else it's too tightly coupled. This would also allow reducing the number of APIs I think, for copy and no_copy variants.

I have to come up with a good ergonomic way to mange it all. It comes at the tradeoff of publicly exposing the pointer wrappers, so more surface for breaking changes in the future. If you want to take a look at it, I wouldn't mind, I can refactor my app after to test if it's ergonomic to use in a real complex app.