분산 시스템 간 공유 변수에 대한 고찰


Memory model (컴퓨터는 released consistency + weak consistency를 통해 동작한다.)


1. Sequential Consistency

- Memory local ordering

- interleaving with other pCPU


2. Released Consistency

- only kept memory ordering with explicit dependency

- if W(A) -> W(B) -> R(A), W(A) -> R(A) ordering is kept, but A and B does not have any dependency relations, which yields instruction level rescheduling

- Thus, we could not say W(A) happens before W(B)

- used for superscalar architecture (instruction level parallelism)


3. Weak Consistency

- Only apply sequential consistency to synchronization variable

- If W(A) -> W(B) -> R(A), and A and B are sync variables, the order is kept

- If W(A) -> W(B) -> W(C) -> R(A), and A and B are sync variables, and C is not sync variable, operation on C can be reordered



Ordering

1. Acquire - Release, When acquire is called in thread A, thread B after release can see all the stored variable

2. Consume - Release, When consume is called in thread A, thread B after release can see only dependent lately stored variable

3. Relaxed, weak consistency


https://en.cppreference.com/w/cpp/atomic/memory_order