Managing Transfer-Based Datacenter Connections

Abstract

Creating datacenter federations allows balancing workloads accommodating spikes in cloud demand while reducing the extensive energy consumption of datacenters. This allows datacenter operators to reduce cost and increase their revenue from using under-utilized resources. To implement those elastic operations, datacenters need to be connected through a network providing huge capacity and guaranteeing adequate resource availability when it is required. Those features can be covered by the flexgrid optical technology provided with an intelligent control plane. For the latter, in this paper we rely on the application-based network operations (ABNO) architecture currently under standardization in the IETF. To increase the availability of the desired capacity, a carrier software defined network (SDN) is placed on top of ABNO. On this architecture, cloud resource managers request data transferences using an application-oriented semantic, which includes data volume and required completion time. Those requests are transformed into connection requests by the carrier SDN. To increase resource availability for incoming requests, the carrier SDN can perform elastic operations on already established connections supporting transferences provided that the committed completion time is ensured. Requests are routed and resource allocation is scheduled, which becomes the routing and scheduled spectrum allocation (RSSA) problem. An integer linear programming model is proposed, and an algorithm for solving the RSSA problem in realistic scenarios is designed. Results showing remarkable gains in the amount of data transported motivate its use for inter-datacenter traffic.