Performance study of an SnF scheduling method for bulk data transfers over inter-datacenter WANs

Abstract

The rise of emerging services has brought a rapid growth in bulk data transfers across inter-datacenter (DC) wide area networks (WANs). To accommodate the demands of such services, DC storage is introduced into the data-plane path so that delay-tolerant bulk data can be temporarily stored and forwarded (SnF) when network is less busy. However, the use of storage transforms a conventional routing problem into a scheduling problem, where spatial routing and temporal scheduling must be jointly performed. Moreover, the complexity of the problem increases exponentially with the scale of the network as well as the number of storage sites. While prior SnF scheduling methods are effective for small networks or for static traffic, they are either too complex or insufficient for large networks with dynamic traffic. In this paper, we study the performance-complexity tradeoff in SnF scheduling. Studies show that desirable performance can be attained by considering a few alternate routes rather than dynamically routing over the entire network topology. Thus, our proposed SnF scheduling method aims to decouple the problem into its spatial and temporal components, and then solve these components separately. The proposed method hence has low complexity while still achieving high performance. Simulations demonstrate that when the traffic load is medium or higher, the conventional joint method may suffer from the detour issue. In this case, the proposed method has the potential to outperform the conventional joint method even with only three alternate routes considered in SnF scheduling.