DeepDefrag: a deep reinforcement learning framework for spectrum defragmentation

Abstract

<p>An exponential growth of bandwidth demand, spurred by emerging network services, often with diverse characteristics and stringent performance requirements, drive the need for more dynamic operation of optical networks, efficient use of spectral resources, and automation. Spectrum fragmentation is one of the main challenges of dynamic, resource-efficient Elastic Optical Networks (EONs). Fragmented, stranded spectrum slots lead to poor resource utilization and increase the blocking probability of incoming service requests. Conventional approaches for Spectrum Defragmentation (SD) apply various criteria to decide when, and which portion of the spectrum to defragment. However, these polices often address only a subset of tasks related to defragmentation, are not adaptable, and have limited automation potential. To address these issues, we propose DeepDefrag, a novel framework based on reinforcement learning that addresses the main aspects of the SD process: determining when to perform defragmentation, which connections to reconfigure, and which part of the spectrum to reallocate them to. DeepDefrag outperforms the well-known Oldest-First FirstFit (OF-FF) defragmentation heuristic, substantially reducing blocking probability and defragmentation overhead.</p>

Deep reinforcement learning for proactive spectrum defragmentation in elastic optical networks

“…This results in almost the same defragmentation overhead as DeepDefrag, enabling an examination of their SBR. On average, the OF-FF (8, 10) and OF-FF (5,15) schemes yield a 20.2% and 29.4% lower SBR than No-SD, respectively, which aligns with the result reported by [9]. As shown in these figures, DeepDefrag has almost the same defragmentation overhead as OF-FF (8,10), while it reduces SBR by 15.8%.…”

“…Two different configurations are evaluated for OF-FF. The first configuration is denoted by OF-FF (5,15), with the SD period equal to 5 connection departures, and allowing up to 15 connection reallocations per SD cycle. OF-FF (5,15) achieves approximately the same SBR as DeepDefrag in the NSFNET topology, allowing for a comparison of their defragmentation overheads.…”

“…The first configuration is denoted by OF-FF (5,15), with the SD period equal to 5 connection departures, and allowing up to 15 connection reallocations per SD cycle. OF-FF (5,15) achieves approximately the same SBR as DeepDefrag in the NSFNET topology, allowing for a comparison of their defragmentation overheads. The second configuration is denoted by OF-FF (8,10), with the SD period equal to 8 request departures and 10 reallocations per cycle.…”

“…To utilize the merit of DRL in automating SD, we proposed DeepDefrag, a novel DRL-based framework that jointly addresses all of the tasks involved in the SD process: determining when to perform defragmentation, which connections to reconfigure, and which part of the spectrum to reallocate them to [15]. DeepDefrag considers the network state to select the most appropriate course of action and can take into account the priorities of a network operator, such as minimizing the number of SD cycles and connection reallocations.…”

“…DeepDefrag considers the network state to select the most appropriate course of action and can take into account the priorities of a network operator, such as minimizing the number of SD cycles and connection reallocations. Our preliminary study in [15], considered only a subset of connections as eligible for reconfiguration and did not examine the spectrum occupancy state in the decision-making process. This paper extends and improves DeepDefrag by (i) considering all connections in the network as candidates for reconfiguration, (ii) considering full information about the spectrum occupancy, including different fragmentation metrics, and (iii) revising the reward function to allow for a more comprehensive evaluation of the impact of actions.…”