Towards Efficient Capacity Planning in Cloud Computing

Dr. Xiaoqiao Meng
IBM T.J. Watson Reserch Center


Abstract

oday's computing cloud achieves an economy of scale by sharing the underlying computing resources (such as CPU, memory and networking bandwidth) among different customers' workload. In modern virtualization-based clouds, capacity planning refers to the strategy of allocating resources to Virtual Machines (VMs) and placement of VMs on physical servers. In this talk, I will present several new VM placement methods for addressing the challenge of dynamic workload in cloud.  These include a method leveraging workload multiplexing, a method reducing network bandwidth usage, and a method solving a probabilistic bin packing problem.