Squeezing More Resources from Compute Cloud via Efficient Capacity Planning

Speaker: Dr. Xiaoqiao Meng

Faculty Host: Professor Jonathan Chao

Abstract

Compute cloud has increasingly become the platform for a large amount of computation and storage tasks that used to be hosted on the Internet or in clusters. Modern compute clouds are based on the virtualization technology and allow applications to share the underlying computing resources (such as CPU, memory and networking bandwidth) by running in isolated virtual machines (VM). A key factor for the economic success of a compute cloud is capacity planning, which refers to the procedure of allocating resources commensurate with VM workload demands and application performance requirement. In this talk, I will introduce several new capacity planning techniques. These include: (1) a technique to leverage the power of workload multiplexing by provisioning multiple VMs as a whole; (2) a network bandwidth-oriented VM placement technique to save network bandwidth usage; (3) a practical method to combine network oriented VM placement with traditional CPU (or memory) oriented placement algorithms.

About the Speaker

Xiaoqiao Meng is a Research Staff Member at the Distributed Computing Department of IBM Thomas J. Watson Research Center. He received his Ph.D. in CS from UCLA in 2006. He received a M.S. from the Institute of Automation, Chinese Academy of Sciences in 2001 and a B.S. from the University of Science and Technology of China. Prior to joining IBM, he was a Research Staff Member at NEC-Labs America. His research interests include cloud computing, wireless networking and performance modeling. He holds 5 patents and has published more than 30 papers in international journals and conferences. He is a member of IEEE and has served as a TPC member or reviewer for various conferences and journals.