feature: Hard Mode HyperNode based binpack scheduling policy needed
What is the problem you're trying to solve
For example , we have a DDP job that needs 4 , and each node requests 8 GPU. The solution2 is bettern than solution 1.
Describe the solution you'd like
The solution2 is better than solution 1.
Additional context
I'd like to do the PR to support hard mode.
/assign @kingeasternsun
Hi, I've also raised a similar issue: https://github.com/volcano-sh/volcano/issues/4368 . My implementation will mainly modify the batch node order, so it shouldn't conflict with yours. Would you mind taking a look when you have time?
Hi, I've also raised a similar issue: #4368 . My implementation will mainly modify the batch node order, so it shouldn't conflict with yours. Would you mind taking a look when you have time?
Is the same problem with yours?