Can someone help me explain the GPU allocation algorithm under the besteffort strategy? #820
Unanswered
gcx-17211270
asked this question in
Q&A
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
https://github.com/NVIDIA/go-gpuallocator/blob/12c3885850231c64f7f67855ff581ebd6e7465bd/gpuallocator/besteffort_policy.go#L33
The Allocate method's comment describes: "it looks for a solution that maximizes the total score when dividing up all available GPUs on the node into sets of size 'size' and then summing their individual scores."
It further explains: "Such a solution is necessary in the general case because of the non-hierarchical nature of the various links that influence the score calculated for each pair of GPUs."
However, I still don't understand why this is necessary. Can someone provide a detailed explanation or give an example to illustrate why failing to do so would result in a suboptimal outcome?
有人可以帮我解释下besteffort策略下的GPU卡分配算法吗?在Allocate方法的注释上描述了:it looks for a solution that maximizes the total score when dividing up all available GPUs on the node into sets of size 'size' and then summing their individual scores.并在其后说明:Such a solution is necessary in the general case because of the non-hierarchical nature of the various links that influence the score calculated for each pair of GPUs. 但是我还是不明白为什么这样是必要的?有人能给我详细说明,或者举个例子说明为什么不这样做就无法得到最优的结果吗?
Beta Was this translation helpful? Give feedback.
All reactions