Achieving the minimum average coflow completion time(CCT) and the isolation guarantees for multi-tenant, is considered a challenge in a cloud environment. This is because the minimum average CCT and isolation guarantees are two conflicting targets, and they cannot be achieved simultaneously. Prior solutions have implemented a single target either minimizing the average CCT or isolation guarantees. The prior solutions are also limited to clairvoyant scheduling. They also assume the availability of the complete knowledge of coflow sizes before the communication starts. In this paper, we propose an efficient scheduling algorithm smallest-height-first DRF(SHFDRF) for near-optimal scheduling and isolation guarantees without prior knowledge of coflow size. SHFDRF achieves the long-term isolation guarantees and the minimum average CCT by the smallest height first and the monopolistic dominant resource fairness bandwidth allocation strategy. The smallest height first and the monopolistic dominant resource fairness bandwidth allocation strategy can also improve link utilization and system throughput. The trace-driven simulation shows that SHFDRF enables communication stages to 1.28\(\times \), 2.27\(\times \), and 6.28\(\times \) faster on the 95th percentile compared to DRF, NCDRF, and Per-Flow Fairness. Even compared with minimum CCT, the completion time of coflow only slowed down by 13.9% on the 95th percentile. Overall, the performance of SHFDRF is acceptable, and it can be applied to the actual datacenter without the limitation of complete prior knowledge.

在云环境中，实现最小平均同流完成时间（CCT）和多租户的隔离保证被认为是一项挑战。这是因为最低平均CCT和隔离保证是两个相互矛盾的目标，因此无法同时实现。先前的解决方案已经实现了单个目标，即最小化了平均CCT或隔离保证。先前的解决方案还限于透视式调度。他们还假定在开始通信之前就可以完全了解同流大小。在本文中，我们提出了一种有效的调度算法，即最小高度优先DRF（SHFDRF），用于在没有同流大小的先验知识的情况下实现接近最佳的调度和隔离保证。SHFDRF以最小的高度优先和垄断性的主导资源公平带宽分配策略来实现长期隔离保证和最小的平均CCT。最小高度优先和垄断性主导资源公平性带宽分配策略也可以提高链路利用率和系统吞吐量。跟踪驱动的仿真表明，SHFDRF使通信阶段达到1.28与DRF，NCDRF和Per-Flow Fairness相比，在第95个百分位数上，\（\ times \），2.27 \（\ times \）和6.28 \（\ times \）更快。即使与最低CCT相比，在第95个百分位数上，coflow的完成时间也仅降低了13.9％。总的来说，SHFDRF的性能是可以接受的，并且可以将其应用于实际的数据中心，而无需完全的先验知识的限制。