Traffic-Aware Resource Controller for IaaS Clouds

Abstract

In cloud data centers, a resource controller is currently used to deploy virtual machines (VMs) and create virtual networks. To decide which servers VMs are deployed on and migrated to, conventional resource controllers take into account the amount of server resources and usage but not physical network loads. Network resources are more difficult to adjust than server resources since manipulations for a VM running on a server affect VMs running on other servers in shared networks. Physical-link traffic is also challenging to regulate at the granularity of a VM because a physical link is used by multiple VMs to transfer their data. To address this challenge, we propose traffic-aware resource scheduler, which leads to dynamically balancing utilization of physical links. To achieve this, the proposed scheduler determines where virtual objects, including VMs and virtual routers, should be deployed or migrated in accordance with subsequent network-traffic derived from periodically accumulated traffic data. To demonstrate the viability, we develop Protea, a traffic-aware resource controller by extending an existing cloud resource controller. We argue that the proposed system improves communications performance between VMs by eliminating traffic congestion in accordance with network flow variation and the configuration change of network architecture.