A Churn-prevented Bandwidth Allocation Algorithm for Dynamic Demands In IaaS Cloud

In current IaaS datacnters, the network bandwidth resource is shared across tenants in a best-effort manner. In this way, tenants are interfered by each other, which leads to unpredictable bandwidth performance and further does harm to the applications deployed in data centers. Besides, with time-varying amount of users to support, tenants choose to adapt to dynamic demands by renting different number of virtual machines(VMs) from IaaS providers at different service stage. Tenants’ dynamic demand for VMs triggers the bandwidth reallocation, which produces bandwidth churn and also contributes the unpredictable performance. Those proposed strategies all put emphasis on VM’s performance isolation by reserving basic bandwidth for VMs. To our best of knowledge, there is no one that has taken the tenants’ dynamic demands for VMs into consideration. In this paper, we model this bandwidth allocation as an optimization problem, which is subject to the basic bandwidth guarantee and bandwidth churn. With the help of the definition of satisfaction degree for host machine, we design a simple and efficient algorithm to solve this optimization problem by using the inequality of arithmetic and geometric means. Under some experiments, we demonstrates our bandwidth allocation strategy can provide not only the predictable performance but also limit bandwidth churn, both of which improve performance of applications deployed in IaaS cloud.