Previous [ 1] [ 2] [ 3] [ 4] [ 5] [ 6] [ 7] [ 8] [ 9] [ 10]


Journal of Information Science and Engineering, Vol. 28 No. 6, pp. 1045-1058 (November 2012)

Per-Client Network Performance Isolation in VDE-Based Cloud Computing Servers*

1School of Electrical Engineering and Computer Science
2Department of Intelligent Convergence Systems
Graduate School of Convergence Science and Technology
Seoul National University
Seoul, 151-742 Korea
E-mail: {jsleel; jhyoo; sshong}
3Digital Media & Communications R&D Center
Samsung Electronics, Co., Ltd.
Suwon-si, 443-803 Korea

In a cloud server where multiple virtual machines owned by different clients are co-hosted, excessive traffic generated by a small group of clients may well jeopardize the quality of service of other clients. It is thus very important to provide per-client network performance isolation in a cloud computing environment. Unfortunately, the existing techniques are not effective enough for a huge cloud computing system since it is difficult to adopt them in a large scale and they often require non-trivial modification to the established network protocols. To overcome such difficulties, we propose per-client network performance isolation using VDE (Virtual Distributed Ethernet) as a base framework. Our approach begins with per-client weight specification and support client- aware fair share scheduling and packet dispatching for both incoming and outgoing traffic. It also provides hierarchical fairness between a client and its virtual machines. Our approach supports full virtualization of a guest OS, wide scale adoption, limited modification to the existing system, low run-time overhead and work-conserving servicing. Our experimental results show the effectiveness of the proposed approach. Every client received at least 99.4% of its bandwidth share as specified by its weight.

Keywords: network performance isolation, cloud computing, virtual distributed Ethernet (VDE), proportionally fair resource allocation

Full Text () Retrieve PDF document (201211_04.pdf)

Received May 31, 2011; accepted March 31, 2012.
Communicated by Junyoung Heo and Tei-Wei Kuo.
* This article is an extended version of the paper that appeared in Proceedings of the 13th International Workshop on Future Trends of Distributed Computing Systems [1]; and it was supported in part by the Technology Innovation Program (No. 10036495) funded by the Ministry of Knowledge Economy (MKE, Korea), by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MEST) (No. 2011-0001572 and No. 2010-0027809), by Digital Media & Communications R&D Center, Samsung Electronics, Co. Ltd. (No. 0421-20110016), by Advanced Institutes of Convergence Technology (AICT), Institute for Green Smart Systems and by Automation and Systems Research Institute (ASRI).
+ Corresponding author.