Open HigashikataZhangsuke opened 1 month ago
The overhead? Does using multiple node will incur latency on the write to same channel on Redis?
It Seems OK if we only do like 2 node VM experiments. (Multiple has same effect, the op is atomic, and also high-efficient) So I do think it's OK to do so.
It seems to be a easy job now. Anti affinity to deploy pods. And then no need for split requests. These requests are pooled, and the CPU mask assigned to each node just reflects the request split!. Just need to do the GC and global scheduling now.
Do it in parallel. When you are tired, you write part of code for multiple node. When you are concerned, you try to fix the bug, find the correct profiling data.