a round robin scheduled serializes requests and assigns then into resources using a round-robin approach. Google round robin. But simply, let’s imagine several buckets numbered 1 to whatever. And many requests. Each request gets put into a bucket. The first request goes into bucket 1. The next into bucket 2. And so on until a request is put into the last bucket. Then it starts putting requests into bucket 1, bucket 2, etc.
That way, there is a poor unfortunate group of soul who gets put into the last bucket even though theyre one of the first groups to try to get a BG.
My proposal doesn’t change the underlying process dramatically. It just tries to place earlier requests into earlier groups first, whenever possible. If there are 100 groups, instead of round robin scheduling all requests over all BGs, it first tries a bunch of BGs early, before going to a later group. I.e., early risers would first be assigned to a bunch of the earlier BGs.
The remainder of the process remains unchanged.