Conference paper Open Access

ECSched: Efficient Container Scheduling on Heterogeneous Clusters

Hu, Yang; Zhou, Huan; de Laat, Cees; Zhao, Zhiming

Operating system (OS) containers are becoming increasingly popular in cloud computing for improving productivity and code porta-bility. 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.

Files (534.8 kB)
Name Size
2018.europar-cameraready.pdf
md5:caf4c5d3c939e06dd024e60424828942
534.8 kB Download
11
38
views
downloads
Views 11
Downloads 38
Data volume 20.3 MB
Unique views 10
Unique downloads 38

Share

Cite as