Operating system (OS) containers are becoming increasingly popular in cloud computing for improving productivity and code portability. However, container scheduling on large heterogeneous cluster is quite challenging. Recent research on cluster scheduling focuses either on scheduling speed to quickly assign resources, or on scheduling quality to improve application performance and cluster utilization. In this paper, we propose ECSched, an efficient container scheduler that can make high-quality and fast placement decisions for concurrent deployment requests on heterogeneous clusters. We map the scheduling problem to a graphic data structure and model it as minimum cost flow problem (MCFP). We implement ECSched based on our cost model, which encodes the deployment requirements of requested containers. In the evaluation, we show that ECSched exceeds the placement quality of existing container schedulers with relatively small overheads, while providing 1.1 × better resource efficiency and 1.3 × lower average container completion time.
CITATION STYLE
Hu, Y., Zhou, H., de Laat, C., & Zhao, Z. (2018). ECSched: Efficient Container Scheduling on Heterogeneous Clusters. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11014 LNCS, pp. 365–377). Springer Verlag. https://doi.org/10.1007/978-3-319-96983-1_26
Mendeley helps you to discover research relevant for your work.