On-Demand Provisioning of Data-Aggregation Sessions Over WDM Optical Networks

We consider efficient network provisioning algorithms for applications that aggregate large data files from multiple remote sites to a central facility (where the aggregated data is further processed). Many important bandwidth-hungry scientific applications use such data aggregation, and it is important to efficiently use network resources to meet their requirements. We term an entire large-scale data-aggregation session as a data-aggregation request (DAR). In this paper, we investigate the problem of on-demand provisioning of DARs over a wavelength-division multiplexing (WDM) backbone network infrastructure. Our DAR provisioning problem is challenging, as for each DAR we need to jointly identify lightpaths (route, assign wavelengths, and groom) for each of the to-be-transferred files, and schedule DAR's file transfers in time. We first model our DAR provisioning problem mathematically as a mixed integer linear program (MILP); to solve our problem in practice, we propose a DAR provisioning algorithm (named DARP). From our numerical results, we find DARP to be efficient when compared with other benchmark algorithms. We study DARP's performance for a varying number of aggregating sites deployed in the network (i.e., sites with supercomputer facilities) and perform a detailed sensitivity analysis on several parameters of our problem. We also investigate the effect of partitioning the data to be transferred into pieces and conclude that, if the partitioning method is carefully designed, slight improvement over the approach that transfers the whole file (DARP) is possible.