Arvados

Currently the keep client starts prefetch of the next block only once the client has started to read from the "current" block. If the client reads the current block much faster than the tranfer time of the block from the server, this degenerates to a bottleneck on the time to fetch one single block at a time.

Test if prefetching multiple blocks ahead of time improves bandwidth utilization. Requires tuning the Keep cache size and the prefetch window, and testing different settings on different clusters.