Fast way to zeroize stable memory?

Developers
1

I was wondering if it is possible to add system API to zeroize 64-bit stable memory?
Something like stable64_zeroize(pages_offset, pages_count).

I’m working on stable collections in ic-stable-memory and for now some of these collections are limited in size because of that. For example a SHashMap requires you to zeroize the whole block of memory, before using it and right now I’m doing it by creating a cleaning vector vec![0u8; SIZE] and writing it into the desired position in stable memory which is both: expensive and limited (since I can’t create a Vec bigger than 4GBs). So the maximum size of a SHashMap is also limited to 4GBs.

There is a workaround - you can write this cleaning vector multiple times to different locations, but I’m afraid this would simply make a canister hit message limit. I believe there should be a lower level (and much cheaper) API to physically discharge bits to zeros that can be exposed through IC’s API, but maybe I’m wrong.

2 Likes
2

Look into Wasm bulk memory operations. The replica should accept them, but your SDK may have trouble generating those bytecodes. Which language are you interested in?

3

Are these bulk operations exposed?

I’m from team rust.

4

I am particularly interested in the type of collections you are building?

6

Stable memory can only be interacted via functions; it’s not exposed as a wasm memory and so wasm bulk memory operations can’t touch it to my knowledge.

7

Ooops, you are right. I missed the “stable memory” part. But not all is lost! @senior.joinu feel free to lobby for (stable memory) system call API equivalents of Wasm bulk memory instructions.

EDIT: I searched and finally found it…
This is the thread where you can voice your wish: Improving the utility of stable memory - #3 by johan