currently, there is not a bandwidth-aware prioritization of origins. If all workers end up pulling from "slow" providers, a replica can spend a long time not efficiently using its resources and being bandwidth constrained. There should be a periodic disconnect of slowest providers when there is contention to allow faster providers to be prioritized.
currently, there is not a bandwidth-aware prioritization of origins. If all workers end up pulling from "slow" providers, a replica can spend a long time not efficiently using its resources and being bandwidth constrained. There should be a periodic disconnect of slowest providers when there is contention to allow faster providers to be prioritized.