Abstract | ||
---|---|---|
As containers run in a distributed manner in clouds, it is important to satisfy network service level objectives (SLOs) for containers. In addition, it has been known that containers utilize more CPU resources for network processing than native processes because of the long networking stack of containers. Thus, for achieving network SLOs, containers require sufficient CPU resources as well as network resources, which we call inter-resource dependency. However, existing cloud schedulers have limitations in that they do not take CPU into account for the network SLO. This paper proposes DepCon that controls CPU resources for containers to satisfy network SLOs. DepCon consists of DepCon scheduler that works in the cloud-level and DepCon agent at each node. We implement DepCon in the most popular container orchestration platform, Kubernetes. Our evaluation results show that DepCon reduces the network performance variance by 20 times while improving the network throughput by 40%. In addition, DepCon reduces the scheduling overhead by 20 times compared to the representative multi-resource scheduling technique like DRF. |
Year | DOI | Venue |
---|---|---|
2021 | 10.1007/978-3-031-06156-1_27 | EURO-PAR 2021: PARALLEL PROCESSING WORKSHOPS |
DocType | Volume | ISSN |
Conference | 13098 | 0302-9743 |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Eunsook Kim | 1 | 0 | 0.68 |
Kyungwoon Lee | 2 | 3 | 2.41 |
Chuck Yoo | 3 | 0 | 0.68 |