Power‐aware routing in networks with quality of services constraints

Abstract

Current network infrastructures are over‐provisioned and thus exhibit poor power efficiency at low traffic load. We consider networks consisting of bundled links, whereby each link has one or more physical cables that can be switched off independently. The problem at hand is to switch off redundant nodes and cables during off‐peak periods, while retaining the quality of services provided to existing traffic demands. Unfortunately, the problem to maximally shutdown redundant nodes and cables is Non‐deterministic Polynomial‐time (NP)‐complete. Henceforth, we design a fast heuristic, called Multiple Paths by Shortest Path First (MSPF), that aims to maximise the number of switched‐off nodes and cables subject to satisfying maximum link utilisation (MLU) and path length (PL) constraints. We have extensively evaluated the performance of MSPF on both real and synthetic topologies and traffic demands. Further, we have compared its performance against two state‐of‐the‐art techniques: GreenTE, usable when each link has one cable, and Fast Greedy Heuristic (FGH), which supports bundled links but only for networks without MLU and PL constraints. On the Sprint network, MSPF can save on average 5% more power as compared to GreenTE while incurring only 1% of GreenTE's running time. While yielding equivalent power savings, MSPF requires only 0.35% of FGH's running time. Finally, setting MLU to at most 50% and PL to no longer than the network diameter, MSPF reduces the power usage of the GÉANT topology by up to 91% for links consisting of 10 cables. For experiments on synthetic topologies with bundle size of 30, MSPF yields a power saving of 89.81%. Copyright © 2014 John Wiley & Sons, Ltd.

Energy management in communication networks: a journey through modeling and optimization glasses

“…x [72] x x [73,74] x x [75,76] x x [77] x Multi-topology routing [78,79] x Energy-Levels [80][81][82][83] x Node expansion, Multiple sleeping states [48,84] x x [85] x x x [86] x x [87,88] x x x x x [89] x [90] x x x x [91] x x Bi-objective function for energy and delay [92] x x x x hop-by-hop routing [93,94] x x x x Two macro-periods [95] x x [96] [97] x [55,98,99] x Redundancy elimination [100,101] x x x x [102] x x x Approximated shared and dedicated protection [103] x x x x [104] x x Failure probabilities x New power state [110] x x [111] x ECMP and robustness [53] Fully splittable [112] free splitting ratio [113,114] [ 115,116] x x Traffic unaware because only link loads considered [117,118] x x OSPF is modified [119] x OSPF is mo...…”

“…In other two articles [73] and [74], the same authors consider another problem variant, according to which nodes can be put to sleep, routing is fully splittable and a pre-computed set of paths is given as input to the model (per-path routing). The formulations presented in [73] and [74] contain several flaws and fail to correctly identify the real problem addressed. The input path set is composed of the K-shortest paths that satisfy each demand d ∈ D. The same scheme already used in many approaches already mentioned is used to put to sleep the elements.…”

“…Every time a cable or a node is tested for sleeping, a heuristic procedure is called to find an alternative feasible routing on the residual network. Note that in [74] the pre-computed paths cannot be longer than a certain threshold, but since paths are provided in input, that is not translated into an additional group of constraints for the corresponding optimization problem.…”

“…The failure probabilities for a demand is related to the failure probability of each single link crossed by the routing path, and by the number of cables which constitute each link itself. Two heuristic algorithms based on the computation of K-shortest paths for each demand (similar to [74]) to put to sleep link cables are presented and tested. Both algorithms are suitable for on-line and off-line implementation.…”

On the performance of online and offline green path establishment techniques

“…Lin et al [41] consider the QoS of flows when switching off network elements. On the other hand, the Green Distributed Routing Protocol for Sleep Coordination (GDRP-PS) algorithm of [42] divides core routers into two sets: traditional routers (TRs) and power saving routers (PSRs).…”