Research Progress and Trend of Coflow Time-Optimal Scheduling in Data Center Network.

Data-intensive applications in data center networks generate a large number of parallel data streams, where the strategy of flow scheduling and the allocation of network bandwidth have become research hotspot issues in this field. Compared with the scheduling of a single data stream, coflow that aims to improve the overall performance of parallel applications can transmit the application layer semantics to the network layer, which is conducive to scheduling decisions by taking full advantage of the application layer semantics. This paper focuses on coflow scheduling with the goal of optimizing completion time, where we review the existing scheduling frameworks and discuss the ideal characteristics in future work. The existing schemes fall into two categories of centralized scheduling and distributed scheduling. Centralized scheduling makes scheduling decisions through the global view of the central scheduler, and distributed scheduling makes scheduling decisions through the local view of adjacent switches. The existing scheduling schemes have made great progress in time optimization, while in-depth research is still needed in the future in terms of fault tolerance, scalability, and starvation avoidance.