Hiroyoshi Kodama

High Efficiency Cloud Data Center Management System Using Live Migration

The footprint of equipment is the actual size a computer device or equipment takes in the room or on the floor of a data center facility. Improvement of the average CPU utilization of data center servers is necessary to offer efficient service with a low footprint. In this paper we present a high efficiency cloud data center management system using live migration. Our system analyzes the power consumption by the rack unit and cooperates with management software of virtual machines (VMs) using application program interface (API) cooperation. We propose a system that shares a data center with an actual use area and a temporary area, and moves VMs of the host server that exceed the supply electric power of the rack in the actual use area to a temporary area. We calculated for a virtual data center with 6000 servers, and observed that the operating ratio of the actual use area should always be kept within 90%. The footprint can be reduced by 40%. The power consumption of the total of the computer room air conditioner (CRAC) and the server can also be reduced by 6.8%. More influence to service of VM movement was also investigated.

Optimizing ICT Equipment via Resource Allocation in Cloud Systems

A business owner using a cloud service desires that the service operated in the cloud system will be operated with as few hardware resources as possible while meeting the service level agreement (SLA). In this paper, we propose a method to clarify the system configuration of the minimum hardware resource that meets the desired service performance (SLA). We conducted an experiment using memcached which is one of cloud services. When changing the number of CPUs of the server executing the memcached application and acquiring the service performance, we found that the latency, which is the number of CPUs and the service performance index, is not a simple proportional relationship. Therefore, in the cloud service, it is important to change the hardware resources of the server executing each application and observe its behavior. We propose a method to perform benchmark comprehensively by changing hardware resources. With this method, it is possible to allocate the minimum hardware resources while meeting the service performance.