Home // International Journal On Advances in Intelligent Systems, volume 7, numbers 1 and 2, 2014 // View article

A Large-scale Power-saving Cloud System with a Distributed-management Scheme

Authors:
Toshiaki Suzuki
Tomoyuki Iijima
Isao Shimokawa
Toshiaki Tarui
Tomohiro Baba
Yasushi Kasugai
Akihiko Takase

Keywords: power saving, QoS, cloud system, virtual-machine migration, distributed management, resource allocation

Abstract:
A large-scale power-saving cloud system with a distributed management scheme is proposed. The system is composed of multiple data centers (DCs) connected by a wide-area network (WAN). In addition, it includes an inter-DC-management server, multiple DC-management servers, and multiple user VM-management servers. To reduce the power consumption of the DCs and the WAN, virtual machines (VMs) are migrated and data-routing paths are optimized under the condition that quality of service (QoS) is maintained by simultaneously providing necessary CPU resources and network bandwidth for services by the VMs. Aiming to enhance our previously proposed system, the management scheme is based on distributed management instead of central management. In the previous system, one DC-management server gathers all information to determine an appropriate alternative server to which VMs are migrated. On the contrary, in the proposed system, to distribute management load, each user VM-management server sends specifications of a VM to be migrated to other user VM-management servers. The other user VM-management servers then independently return a list of alternative servers that can accommodate the intended VM. After receiving the lists, the user VM-management server selects the most-suitable server. A prototype of the proposed system comprising 1,000 VMs, 400 servers, and four DCs was developed and evaluated. The time for determining reallocation of 1,000 VMs is within five minutes, which is about five times shorter than that taken by the previous system. These results indicate that the proposed system can reduce power consumption for one-week cloud operation by 30%.

Pages: 326 to 336

Copyright: Copyright (c) to authors, 2014. Used with permission.

Publication date: June 30, 2014

Published in: journal

ISSN: 1942-2679