Decentralized task-aware scheduling for data center networks 论文

摘要

Many data center applications perform rich and complex tasks (e.g., executing a search query or generating a user's news-feed). From a network perspective, these tasks typically comprise multiple flows, which traverse different parts of the network at potentially different times. Most network resource allocation schemes, however, treat all these flows in isolation -- rather than as part of a task -- and therefore only optimize flow-level metrics.