No abstract

The unprecedented advances in Passive Optical Networks (PONs) and their proven performance in access networks have encouraged researcher to investigate the use of passive optics to address the limitations of conventional data centre architectures. In this paper, we present a scalable, high capacity and energy efficient arrayed wave guide grating router (AWGR) PON based data centre architecture to facilitate inter and intra racks communication within data centres. The proposed architecture can be scaled up efficiently to hundreds of thousands of servers and has shown energy savings of 45% and 80% compared to the Fat-Tree and BCube architectures, respectively.

Over the last decade, the evolution of data center architecture designs has been mainly driven by the ever increasing bandwidth demands, high power consumption and cost. With all these in mind, a significant potential to improve bandwidth capacity and reduce power consumption and cost can be achieved by introducing PONs in the design of the networking fabric infrastructure in data centers. This work presents a novel server-centric PON design for future cloud data center architecture. We avoided the use of power hungry devices such as switches and tuneable lasers and encouraged the use of low power passive optical backplanes and PONs to facilitate intra and inter rack communication.We also tackle the problem of resource provisioning optimization and present our MILP model results for energy efficient routing and resource provisioning within the PON cell. We optimized the selection of hosting servers, routing paths and relay servers to achieve efficient resource utilization reaching 95% and optimum saving in energy consumption reaching 59%.

Abstract-Recent years have witnessed an ever-increasing growth for cloud computing services and applications housed by data centers. PON based optical interconnects for data center networks is a promising technology to offer high bandwidth, efficient utilization of resources, reduced latency and reduced energy consumption compared to current data center networks based on electronic switches. This paper presents our proposed scheme for data center interconnection to manage intra/inter communication traffic based on readily available low cost and power PON components. In this work, we tackle the problem of resource provisioning optimization for cloud applications in our proposed PON data centers architecture. We use Mixed Integer Linear Programming (MILP) to optimize the power consumption and delay for different cloud applications. The results show that delay can be decreased by 62% for delay-sensitive applications and power consumption can be decreased by 22% for non-delay sensitive applications.

In this paper, we present three heuristics for mitigating post-disaster traffic floods. First exploiting the excess capacity, second rerouting backup paths, finally redistributing the whole traffic by rerouting the working and protection paths to accommodate more floods. Using these mitigation approaches can reduce the blocking by up to 30%.