Despite being efficient approaches to reduce the energy consumption, most of the aforementioned approaches are still limited since they are off-line. Assuming accurate traffic patterns fixed and known a priori may not be appropriate for current dynamic networks in which users can join or leave the network in an unpredictable way, affecting the overall traffic. Evidently, an online input is a more realistic consideration for energy-aware solutions and allows to dynamically adapt the number of active network elements to the arriving traffic.
2.4. Dynamic Green Traffic Allocation
energy consumption, determined as the number of active Open-Flow switches in the network. For this, a low complexity algorithm is presented using, for each pair of endpoints, a pre-computed set of shortest paths to select the route that minimizes the number of switches that become active after allocating the flow. Although this proposal allows real-time operation routing flows sequentially, only low-loaded nighttime traffic is considered, failing to extensively examine the implications of more demanding scenarios.
A similar approach is conceived in [102], where a dynamic routing scheme applying traffic aggregation for each incoming flow is proposed. Instead of only considering the number of hops, in this proposal the number of active links and nodes is also used as a routing metric. Paths are only taken into account if they have sufficient residual bandwidth to assume the incoming flow. In case of congested shortest paths, the path minimizing the MLU is selected.
In order to deal with traffic change in real-time manner, a Centralized Energy-efficient Rout- ing Control (CERC) strategy is proposed in [103]. In this strategy the centralized controller is in charge of four main functions: link status monitoring, link sleeping, link awakening, and link status forecast. In essence, during traffic idle times low-loaded links are put into sleep mode and switched back on once the traffic increases, in order to avoid network congestion. During the sleeping procedure, candidate links are selected following two criteria of link utilization: traffic amount and number of node pairs. Routing paths are calculated based on the link metrics and the use of Equal Cost Multiple Path (ECMP). Additionally, the granularity of interval time for uploading the link status information to the controller is analyzed in this paper. However, simu- lations only consider a synthetic topology, without evaluating the performance of such approach in real-world networks with measured traffic traces.
The authors of [104] present the design of an Energy Monitoring and Management Applica- tion (EMMA) to minimize energy consumption in SDN-based backhaul networks. They formu- lated this problem as a non-linear optimization model and proposed heuristic algorithms for the dynamic routing of flows and the management of the resulting link and switch activity. However, such algorithms were implemented in an SDN emulation environment with out-of-band control traffic, limiting their applicability to networks where dedicated links between the controller and forwarding nodes are deployed. Different proof-of-concept prototypes showing the applicability of EMMA in three realistic use cases (i.e. a software switch network, a mmWave mesh network and an analogue RoF domain for ground-to-train radio access) are discussed in [105].
An online flow-based approach that takes into account the dynamic arrival and departure of users in SDN-based campus networks is designed in [106]. In this paper, the authors formulate the problem of routing the new incoming flow and dynamically re-optimizing the existing flows as an ILP subject to QoS constraints (i.e. bandwidth and delay), which aims to reduce the total energy consumption in the wireless and wired parts of the network. Given the NP-hard nature of such problem, an ant colony-based heuristic, called Ant Colony Online Flow-based Energy-efficient Routing (AC-OFER), is proposed to approximate to the ILP optimal solution. Considering the undesired consequences of recomputing the routing paths in dynamic energy saving approaches, in [107] an energy efficient routing scheme based on Fast Rerouting (FRR), namely GreenFRR, is introduced. This paper aims to reduce the routing convergence time considering the occurrence of frequent traffic changes in a network. To do so, authors first formalize the FRR-based energy efficient routing problem and prove the associated NP-hardness. Thus, heuristic algorithms are proposed, which maximize the number of sleeping links and provide available rerouting paths quickly when a routing convergence is triggered.
In [46] authors propose ElasticTree, a network-wide power manager to save energy in data centers using SDN. This solution dynamically finds the minimum set of network elements re- quired by changing traffic loads, while satisfying performance and fault tolerance constraints. In this regard, three strategies were studied, namely Formal Model, Greedy Bin-Packing and Topology-aware Heuristic. While the first option presents scalability issues and the second saves less power, the best performance is obtained by the Topology-aware Heuristic. However, this approach is specifically conceived for FatTree networks.
Another approach about power efficiency in software defined data center networks is pre- sented in [108]. In this work different energy-aware routing strategies, combining common rout- ing and scheduling algorithms, are evaluated and implemented as a OpenNaaS-based prototype. However, these strategies are only applicable in data centers and are also incompatible with environments without dedicated control networks.