دانلود رایگان مقاله قرار دادن محتوای مشترک و مسیریابی lightpath در تخصیص طیف در CDN

Abstract

In this work, we address the problem of jointly deciding the placement of the contents delivered by a Content Distribution Network (CDN) among the available data centers, together with the allocation of the lightpaths required to serve the anycast demands initiated by the CDN network users, assuming an underlying high-capacity Elastic Optical Network (EON). We firstly present an Integer Linear Programming (ILP) formulation to optimally solve the targeted problem. This ILP formulation is of high complexity, though, and cannot be used to solve realistically sized problem instances. Hence, we also introduce a novel heuristic called CPRMSA-PD, which decomposes the problem into three sub-problems and applies greedy heuristics and simulated annealing meta-heuristic techniques to yield accurate solutions with practical execution times. We validate the performance of our CPRMSA-PD heuristic in medium-sized problem instances by comparing its results to the ones of the optimal ILP formulation. Next, we use it to give extensive insights into the effects of different key parameters identified in large CDN over EON backbone networks.

7. Concluding remarks

This paper has addressed the joint Content Placement and lightpath Routing Modulation format and Spectrum Assignment (CPRMSA) problem in Content Distribution Network (CDN) over Elastic Optical Network (EON) scenarios. To solve it, an optimal ILP formulation (CPRMSA-ILP) as well as a heuristic based on problem decomposition (CPRMSA-PD) have been proposed. Once the performance of CPRMSA-PD has been successfully validated against the optimal CPRMSA-ILP model in medium-sized problem instances, we use it to analyze the effects that key parameters detected in the targeted scenario (number of DCs, replica server capacity, number of content groups, and content group popularity distribution function) can have on the utilization of the underlying EON resources. From the obtained results, sufficient capacity to allow content replication in the DCs yields dramatic reductions in terms of frequency slices used, allowing for superior anycast demand aggregation, together with the usage of advanced modulation formats (PM-8-QAM, PM-16-QAM instead of PM-BPSK or PM-QPSK). Furthermore, while no substantial effects of the number of content groups on the required number of frequency slices are observed, we identify potential EON underutilization in highly-skewed content group popularity distributions if very limited content replication prevents effective demand aggregation. It should be noted that the addressed optimization problem is static and assumes that popularity of Content Groups is fixed. However, the proposed approach could be also applied to such scenarios that take evolving popularity trends into account. For example, the proposed static optimization process could be repeated periodically (or triggered upon significant CG popularity changes) using the new data at the moment. Next, reconfiguration to the new optimal setup would follow. Alternatively, the proposed static optimization process could only account for CGs whose popularity is expected to remain quite constant over time. Then, for those CGs whose popularity is expected to show higher variability over time, as well as for the associated anycast demands, lighter dynamic CP and lightpath RMSA heuristics and procedures could be employed. Both applications are identified as possible research lines and left for future work at this point.