[Week 7]: A Partial Stable IPC Bench

[lvyuemeng]

Summary

PR

• A reduction of boilerplate by type nomination.

PR

• A interface generalizaiton by ISender/IReceiver api sealed.

PR

• A bench attempt from 张超(loichyan), implement evering/shimipc/monoio.
• copy_from_slice of ShmBox<[T]>.

PR

• Add ipc test for bench: char_ipc/box_ipc.
• Refactor ShmToken by erasing ?Sized type for Send.
• [Deprecated]: few modification of IdCell, bench test for validation.

PR

• refactor double free of unmap.
• refactor incorrect read rather take out of OpCache.
• add (partial) correct test code of evering.

Bench

Environment

wsl2 ubuntu 24

---

Code

The general pattern is to mock below workflow:

• client/server initiate shared memory in each isolated thread.
• client send request + request data: spawn CONCURRENCY = 200 with each iters = total_iters/CONCURRENCY; generate request data from bufsize.
• server receive request and request data, verify request and modify request data and send back.
• client receive response and verify.
• calculate the total elapsed from client sending to correctly receiving for all spawned threads.

PR:

• shmipc/monoio: pr
• evering: pr

Result

It will crash after a few times of sample test, thus I repeat multiple times and list one of above. The time ratio is akin in multiple times like above for sure.

Problem

Due to somehow incorrect design(unfree/double free), it will crash after a few times of tests and the benchmark didn't successfully finish, thus only above results could be shown, however, I try multiple times to test and the time ratio is similar, which could partially verify in certain environment.

The ergonomics is bad due to previous incorrect design, I would pause while to read more.