Time-Space Decoupled SnF Scheduling of Bulk Transfers Across Inter-Datacenter Optical Networks

Abstract

To meet the increasing inter-datacenter traffic, datacenter storage is brought in the forwarding path. Bulk data that are delay-tolerant can be temporarily stored and forwarded (SnF) at a later time when link is less busy. However, the use of storage transforms the conventional routing problem into a scheduling problem, where both bandwidth and storage resources must be allocated and both spatial routing and temporal scheduling must be performed. Such an SnF scheduling problem is critically important for the efficiency of SnF approaches. Most prior solutions aimed to jointly solve its temporal and spatial components and formulated this problem into difficult optimization problems, which contributes to a huge expansion size of the problem and hence are too complex for large networks and dynamic traffic. In this paper, we present analytic models to quantify the performance-complexity tradeoff in the SnF scheduling problem. Our key findings reveal that desirable performance can be obtained by considering only a few pre-selected routes rather than searching in the entire network topology. Thus, we propose a time-space decoupled (TSD) SnF scheduling method. Compared to the conventional joint methods, the advantages of the TSD method are as follows: (i) by decoupling the problem and solving them separately, the TSD method reduces the quadratic complexity of the joint methods to linear complexity; (ii) by condensing the redundant states, the TSD method obtains a longer horizon of temporal scheduling, given the same computational cost; (iii) by bounding the spatial hop count of routing paths, the TSD method avoids the detour issue faced by the joint methods and hence uses bandwidth more efficiently; (iv) by formulating the problem into a routing problem, the TSD method greatly simplifies the problem for dynamic traffic. Simulations demonstrate that the TSD method can outperform the conventional joint method, especially when the traffic load is moderate-to-high.