On the Parallelization of Spectrum Defragmentation Reconfigurations in Elastic Optical Networks

• Published in: IEEE/ACM transactions on networking Vol. 24; no. 5; pp. 2819 - 2833
• Main Authors: Zhang, Mingyang, You, Changsheng, Zhu, Zuqing
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.10.2016; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithm design and analysis; Bandwidth; Elastic optical networks (EONs); Lagrangian relaxation; Optical fiber networks; Optical switches; parallel defragmentation; Routing; SONET; Spectrum allocation; spectrum defragmentation; WDM networks
• ISSN: 1063-6692; 1558-2566
• DOI: 10.1109/TNET.2015.2487366

Flexible-grid elastic optical networks (EONs) have attracted intensive research interests for the agile spectrum management in the optical layer. Meanwhile, due to the relatively small spectrum allocation granularity, spectrum fragmentation has been commonly recognized as one of the key factors that can deteriorate the performance of EONs. To alleviate spectrum fragmentation, various defragmentation (DF) schemes have been considered to consolidate spectrum utilization in EONs through connection reconfigurations. However, most of the previous approaches operate in the sequential manner (Seq-DF), i.e., involving a sequence of reconfigurations to progressively migrate highly fragmented spectrum utilization to consolidated state. In this paper, we propose to perform the DF operations in a parallel manner (Par-DF), i.e., conducting all the DF-related connection reconfigurations simultaneously. We first provide a detailed analysis on the latency and disruption of Seq-DF and Par-DF in EONs, and highlight the benefits of Par-DF. Then, we study two types of Par-DF approaches in EONs, i.e., reactive Par-DF (re-Par-DF) and proactive Par-DF (pro-Par-DF). We perform hardness analysis on them, and prove that the problem of re-Par-DF is NP-hard in the strong sense while pro-Par-DF is an APX-hard problem. Next, we focus on pro-Par-DF and propose a Lagrangian-relaxation (LR) based heuristic to solve it time-efficiently. The proposed algorithm decomposes the original problem into several independent subproblems and ensures that each of them can be solved efficiently. The LR based approach informs us the proximity of current feasible solution to the optimal one constantly, and offers a near-optimal performance (relative dual gap 5%) within 500 iterations in most simulations. Extensive simulations also verify that the proposed pro-Par-DF approach outperforms Seq-DF in terms of the DF Latency, Disruption and Cost.