1.4 Thesis Organization
2.1.1 Traffic-Based Solutions
Under the assumption of dealing with expected (i.e., known in advance) traffic, traffic-based solutions are routing mechanisms that aggregate traffic over a network subset in over-provisioned systems. By adopting this strategy, the number of turned on network components handling the incoming traffic is minimized. Over the last decade, traffic-based solutions have been widely studied in order to tackle the problem of power consumption.
One of the most significant works in this area is [53], in which authors propose GreenTE, an intra-domain, centralized TE mechanism that finds a set of links that can be turned off under
2.1. Principles of Energy-Aware Routing
a given traffic load or matrix. The approach is based on a Mixed-Integer Linear Programming (MILP) formulation where the traffic demands are routed through a set of previously computed k-shortest paths. Performance requirements such as Maximum Link Utilization (MLU) and network delay are considered as constraints in the problem.
In a similar way, Bianzino et al. [54] aim to find the network configuration that minimizes the network energy consumption, modeled as the sum of the energy spent by all nodes and links carrying traffic. To achieve this, they formulated an optimization problem for finding minimum-power network subsets assuming the existence of traffic level with known daily be- haviour. Therefore, an accurate prediction of incoming traffic is required.
In [55] the problem of switching off network elements is formulated as a variant of the Multi- Commodity Minimum Cost Flow (MCMF) problem [56]. A greedy heuristic that iterates first over all the network nodes and then through the links (both sets sorted according to rules such as random, least-link, least-flow and most-power) is proposed. The authors studied all possible node/link sorting combinations.
More recently, in [57], the authors introduce a state-of-the-art study of energy efficiency strategies in SDN. This paper addresses the importance of implementing green routing methods in SDN, taking advantage of the flexibility given by dynamic configuration and centralized network view capabilities. A summary of some existing energy-aware techniques in SDN with their key properties (benefits and drawbacks) is presented, based on a four category classification namely: traffic aware, compacting TCAM, rule placement, and end host aware.
Regarding the use of partially deployed SDN, the authors of [58] face the problem of saving energy in these hybrid scenarios. For this, they formulate an optimization model which aims to find minimum power network subsets. After proving the problem is NP-hard, they propose a heuristic solution to approach the exact solution, based on the use of several groups of spanning trees to satisfy the traffic loads.
The energy efficiency in hybrid IP/SDN networks is also addressed in [59]. In particular, this paper introduces the energy-aware SDN nodes replacement problem aiming to improve the energy efficiency during the transition from traditional IP networks to fully deployed SDN. To solve this problem an Integer Linear Programming (ILP) formulation and a genetic algorithm are proposed. The most appropriate set of traditional IP nodes to be upgraded to SDN-enabled switches are selected according to six different replacement methods.
Taking into account the rule space limitations of the TCAM in SDN forwarding nodes, Giroire et al. [60] propose an energy-aware routing method for a backbone network. An ILP optimization model is presented as well as an efficient heuristic, respecting capacity constraints on links and rule space constraints on routers. Performed simulations show that important savings (similar to the classical TCAM-agnostic approach) can be achieved using the proposed smart rule space allocation.
Markiewicz et al. [61] formulated an MILP model that aims to switch on a minimum amount of routers and links to carry the traffic. To solve the problem for large networks, they present a heuristic method, called Strategic Greedy Heuristic (SGH), that iteratively selects a pre- computed shortest path for each request, according to four different strategies of processing order of requests.
Similarly, two heuristic algorithms, namely Next Shortest Path (NSP) and Next Maximum Utility (NMU), are proposed in [62] to deal with the energy aware routing problem in SDN. Considering an initial Shortest Path Routing (SPR), both models seek to minimize the energy consumption of links and switches redirecting selected flows from under-utilized links to a more utilized replacement path in order to turn off the under-utilized links if all flows are redirected. The new selected path corresponds, respectively, to the next shortest path or the most-loaded path calculated after excluding the under-utilized links. Results show that, although NSP and NMU are more efficient in minimizing the average path length, more energy can be saved initializing the network with the outputs of [61] instead of using the SPR approach.
The authors of [63] provide two greedy algorithms for minimizing the power of integrated chassis and line-cards used under constraints of link utilization and packet delay. One attempts to adjust as few requests as possible while the other reroutes all requests sorted by priority in order to get better energy savings. To achieve this they considered an expanded network topology according to the connections between forwarding nodes. Specifically, routers are symbolized as star graphs, where the center represents the integrated chassis and the leaves are the line-cards. Although the proposed scheme saves an important amount of energy, it results in a highly-loaded network environment, making the network vulnerable to link failures and sudden traffic bursts. Focusing on the use of pre-established multi-paths, an SDN-based green routing and resource management model for Multi-Protocol Label Switching (MPLS) networks is presented in [64]. In this approach, the controller, considering several pre-established Label Switching Paths (LSPs)
2.1. Principles of Energy-Aware Routing
between each ingress and egress pair, performs three main operations: path selection, load balancing and path resizing. In the first case, paths are activated or deactivated in order to reduce energy consumption. The two other functions are intended to manage resource utilization. Instead of assuming dedicated links between the controller and SDN nodes, in [65] the authors propose a model for controller-switch associations that aims to maximize the energy efficiency of the network. Although the routing of control traffic is considered in this work, they assume that controllers act as well as forwarding nodes, i.e. data plane traffic demands are routed through network controllers. Therefore, only links that belong to control paths are activated and data traffic demands are routed using these links until an MLU threshold is reached. We argue that data plane traffic should not pass through network controllers, since this will represent an additional load in these devices.