CATS: Cloud application-specific traffic scheduling and routing

Abstract

This thesis addresses the limitations and challenges faced by traditional networks via layering, with a focus on the cloud environment that a large fraction of applications and services are running on today. We separate the cloud networks into two categories: intra cloud networks and inter cloud networks. For intra cloud networks, we show a network resource management layer can achieve application-aware networking and improve end-to-end application performance. For inter cloud network, we show an overlay network built on top of the cloud platform can consistently improve the network performance of file transfer for a large fraction of users. Two middleware systems are presented in the thesis: Phurti and CRONets(Cloud-Routed Overlay Networks). Phurti improves MapReduce job completion time for 95% of the jobs and decreases average job completion time by 20%, while CRONets improves 78% of the default Internet paths with a median improvement factor of 3.27 and 1.67 times for more than 6,600 Internet paths examined.