Time-Saving First: Coflow Scheduling for Datacenter Networks

• Published in: 2017 IEEE 86th Vehicular Technology Conference (VTC-Fall) pp. 1 - 5
• Main Authors: Borjigin, Wuyunzhaola, Ota, Kaoru, Dong, Mianxiong
• Format: Conference Proceeding
• Language: English; Japanese
• Published: IEEE 01.09.2017
• Subjects: Bandwidth; Mathematical model; Optimization; Schedules; Scheduling; Servers; Time factors
• DOI: 10.1109/VTCFall.2017.8288339

Coflow is a collection of parallel flows, while a job consists of a set of coflows. A job is completed if all of the flows completes in the coflows. Therefore, the completion time of a job is affected by the latest flows in the coflows. To guarantee the job completion time and service performance, the job deadline and the dependency of coflows needs to be considered in the scheduling process. However, most existing methods ignore the dependency of coflows which is important to guarantee the job completion. In this paper, we take the dependency of coflows into consideration. To guarantee job completion for performance, we formulate a deadline and dependency-based model called MTF scheduler model. The purpose of MTF model is to minimize the overall completion time with the constraints of deadline and network capacity. Accordingly, we propose our method to schedule dependent coflows. Especially, we consider the dependent coflows as an entirety and propose a valuable coflow scheduling first MTF algorithm. We conduct extensive simulations to evaluate MTF method which outperforms the conventional short job first method as well as guarantees the job deadline.