On IGP link weight optimization for joint energy efficiency and load balancing improvement☆
Introduction
Computer networks have been consuming an increasing amount of energy because of their greater pervasiveness due to the need to support new network applications such as video based services and cloud computing. Nowadays, network operators are keen to identify new efficient ways of greening their networks so as to reduce both their operational costs and environmental impact. By 2020, European Telecoms are expected to consume around 35.8 TW h, a rapid rise from the current 21.4 TW h consumed per year, if greener networks are not deployed [1]. More specifically, the share of backbone networks in the total network power consumption will increase from 10% to 40% by 2020 if no green actions are taken. This rapid change is mainly driven by the ever increasing popularity of bandwidth-hungry applications over the Internet. In order to tackle the rapid increase in the energy consumption of backbone networks, different Energy-aware Traffic Engineering (ETE) schemes have been proposed in the last few years [2], [3], [8], [9], [11], [12]. These schemes apply different sleep reconfiguration and transmission rate adaptation techniques at network devices to achieve enhanced energy efficiency. One typical strategy is to use only a subset of network devices for carrying customer traffic when the traffic volume is low during off-peak time. This allows other network devices to have the opportunity to save energy, e.g. through sleeping. As long as the reduced network capacity is sufficient for handling the traffic demand, energy can be saved without causing service deterioration to end users.
On the other hand, network operators traditionally employ traffic engineering (TE) schemes [4] for the sole purpose of load-balancing because this was their main concern in the past before the energy efficiency issue also became a concern. Traditionally, the ultimate objective of load-balancing is to reduce the Maximum Link Utilization (MLU) in the network through optimized distribution of traffic. This reduction in MLU allows the networks to offer better Quality-of-Service (QoS) assurance and also to efficiently handle unexpected traffic surges. However, conventional load-balancing and ETE are intuitively conflicting with each other in network configurations since load-balancing attempts to distribute the traffic as evenly as possible while ETE algorithms attempt to concentrate traffic on the smallest feasible subset of active devices (e.g. network links) in order to allow other elements to sleep. To intelligently resolve such a tussle, a new ETE scheme called Green Load-balancing Algorithm (GLA) is proposed in this paper. The novelty of such a scheme is that it jointly optimizes the load-balancing and energy efficiency in a holistic manner. GLA achieves such objectives by optimizing the Interior Gateway Protocol (IGP) link weights in order to maximize energy saving gains through link sleeping, while maintaining, or even further improving the load-balancing performance of the residual working topology. This is in salient contrast to the conventional single-objective IGP link weight setting schemes that optimize for load-balancing only and therefore do not efficiently provide opportunities for link sleeping operations. In summary, our contribution is a practical solution that opens a new dimension of energy efficiency optimization, but without sacrificing traditional traffic engineering performance in plain IP routing environments.
The rest of this paper is organized as follows: the related work is described in detail in Section 2 and an illustrative example to highlight the concepts and reasoning of GLA is given in Section 3. In Section 4, the problem formulation of optimizing link weights for both load-balancing and energy efficiency is presented. A short description of the three ETE schemes that are used in this paper to evaluate the performance of GLA is also provided in Section 4. An in-depth description of the GLA mechanism is given in Section 5 and GLA is evaluated with the help of the Point-of-Presence representation of the European academic network GÉANT and its real traffic matrices in Section 6. GLA is shown to substantially improve both the load-balancing and energy efficiency of the three chosen existing ETE schemes. In Section 7, GLA is further customized for a specific ETE scheme and is shown to improve even more the performance of this ETE scheme. This shows that GLA can be regarded as a generic enough technique that can be further customized based on the specifics and particularities of the ETE schemes it is applied on. In Section 8, the maximum propagation delay of the different ETE schemes is evaluated. Since, as will be shown, this delay can be excessive as a result of their operations, the three existing ETE schemes are modified in Section 9 so that they can respect a constraint on the maximum propagation delay. The new delay-aware ETE schemes are evaluated against their original counterparts in Section 10. Finally, the paper is summarized and some directions for future work are provided in Section 11.
Section snippets
Related work
A number of strategies are currently being followed to reduce the energy consumption of backbone networks. The first strategy is the use of more energy-efficient hardware in network equipment. Energy efficiency of the hardware can be improved by employing lower power circuit design at the physical level [9], [10], [11] and using dynamic frequency and voltage scaling at the functional level [12], [13], [14]. Secondly, it is also possible to make use of greener networking protocols. For example,
An illustrative example of GLA
In order to illustrate the basic concept of GLA-based link weight optimization for both load-balancing and energy efficiency, we use the small example network topology in Fig. 1 with the indicated link capacities and IGP weight settings. The aim of such an example is to illustrate how IGP link weights can be manipulated in order to create opportunities for more links to sleep, but without affecting the load-balancing requirements. For simplicity and clarity, we use an “incomplete”
Problem formulation
The joint-optimization of load-balancing and energy efficiency in a network can be expressed with the following two objectives (see Table 1):subject to:
Eq. (1) represents the first objective of GLA which is the minimization of the Maximum Link Utilization (MLU) in the network in order to achieve load-balancing. For the GLA scenarios based on one traffic matrix, it refers to the MLU related
Scheme overview
Since it is well-known that computing the optimal link weights for basic load-balancing alone is already an NP-hard problem [35], here we propose a new scheme, Green Load-balancing Algorithm (GLA), which is based on meta-heuristics (evolutionary/genetic algorithms) to find the optimized IGP link weights which can solve the more complicated problem of the joint-optimization of load-balancing and energy efficiency in a backbone network.
More specifically, GLA is used to solve the problem of
Network scenario
We evaluate the performance of GLA on top of the three different ETE schemes, LF [21], MP [22] and TLS [7], [8], by using the operational network topology, GÉANT and its published traffic matrices [37]. GÉANT is a European academic network which has allowed researchers access to its network topology and traffic matrices. The published topology consists of 23 Points-of-Presence (PoPs) and 74 unidirectional links of varying bandwidth capacities which are described in Table 2 below. The total
Solution-enhancement heuristic
The performance of GLA for TLS can be improved further if GLA is further customized for TLS through the use a solution-enhancement heuristic (SH). This enhanced version of GLA is called Green Load-balancing Algorithm with solution-enhancement heuristic (GLA-SH). SH operates on the best solution in the population at the end of each iteration of GLA. The best solution is determined by a single aggregated objective function represented by the ratio of Maximum Link Utilization to energy efficiency
Packet delay performance of the existing ETE schemes
The end-to-end maximum packet delay (MPD) of operational networks has become increasingly important due the popularity of real-time applications such as Voice-over-IP (VoIP) and video streaming [39]. These applications normally require a bounded MPD in order to provide end-users with good Quality-of-Experience (QoE). The three ETE schemes, described in Section 4.2: LF, MP and TLS, are oblivious to packet delay and therefore, the packet delay may substantially increase when links are put to
Improvement to current ETE schemes for maximum packet delay constraint
In the previous section, it was indicated that the MPD performance of the three plain ETE schemes can vary substantially with the three sets of link weights. The current ETE schemes do not have a deterministic way of ensuring that a bounded MPD performance is maintained. Therefore, there is no guarantee that the MPD performance will remain within the tolerance limits of network operators. In light of this observation, it is desired to further extend the current ETE schemes so as to guarantee a
Performance evaluation of ETE schemes with maximum packet delay constraint
We evaluate the three new MPD-aware ETE schemes, namely LF-D, MP-D and TLS-D, by using the same network scenario as described in Section 6.1 and compare the load-balancing and energy-savings performance when the Default and GLA link weights are used. The IGP-WO link weights were not used during the evaluation because they give an MPD value which is very high. Specifically, the newly introduced extra delay due to the optimization of the link weights for load-balancing only is substantially
Summary
In this paper, we proposed the Green Load-balancing Algorithm (GLA), which intelligently optimizes IGP link weights in IP backbone networks in order to improve both the load-balancing and energy efficiency of existing ETE schemes. GLA uses a customized multi-objective genetic algorithm to identify the optimized solutions. In addition, two new custom mutation and crossover operators have been designed to improve the performance of the genetic algorithm by enabling the solution space to be
References (39)
- et al.
Analysis of point-to-point packet delay in an operational network
Comput. Netw.
(2007) - et al.
The potential impact of green technologies in next-generation wireline networks: is there room for energy saving optimization?
IEEE Commun. Mag.
(2011) - et al.
A survey of green networking research
IEEE Commun. Surv. Tutor.
(2010) - et al.
Energy efficiency in the future internet: a survey of existing approaches and trends in energy-aware fixed network infrastructures
IEEE Commun. Surv. Tutor.
(2010) - et al.
An overview of routing optimization for internet traffic engineering
IEEE Commun. Surv. Tutor.
(2008) - et al.
Energy-minimized design for IP over WDM networks
IEEE J. Opt. Commun. Netw.
(2009) - M. Zhang, C. Yi, B. Liu, B. Zhang, GreenTE: Power-Aware Traffic Engineering, in: Proc. 2010 of IEEE International...
- F. Francois, N. Wang, K. Moessner, S. Georgoulas, Optimization for time-driven link sleeping reconfigurations in ISP...
- et al.
Optimizing link sleeping reconfigurations in ISP networks with off-peak time failure protection
IEEE Trans. Netw. Serv. Manage.
(2012) A radical new router
IEEE Spectr.
(2009)
Enabling backbone networks to sleep
IEEE Netw.
Cited by (14)
Energy management in communication networks: a journey through modeling and optimization glasses
2016, Computer CommunicationsCitation Excerpt :Note that, as in Section 4.2, the analyzed approaches are presented in multiple sub-sections according to the most relevant problem features. A genetic algorithm to solve in a centralized offline manner a joint optimization problem to minimize both network consumption and maximum link utilization is presented in [126] and [127]. Although the authors claim to solve a classic EANM problem with multicommodity flow (flow-based routing), the proposed procedure is based on the optimization of the link weights of an Interior Gateway Protocol: for this reason we place this work in the shortest-path routing section.
Low-Carb: A practical scheme for improving energy efficiency in cellular networks
2016, Computer CommunicationsCitation Excerpt :While backbone networks are not the largest energy consuming segment of telecommunication networks, their share of power consumption is expected to grow due to increasing traffic demands. Recent examples of work focusing on energy efficient backbone networks include [96–100]. Instead of the all-or-half approach of Fig. 4(a), power-saving may also be applied in a more granular way, such as that shown in Fig. 4(b), whereby one-third of the TRX may be switched on/off independently in response to current traffic volume.
Minimization of network power consumption with redundancy elimination
2015, Computer CommunicationsCitation Excerpt :For instance, as shown in [22], only few configurations (two or three per day) are enough to significantly save energy. Similarly, the authors in [23] suggest to use only two network configurations for a daily time. During peak hours, all the links in the network are active while EAR (with sleep links) is considered only in off-peak time to save energy.
Network routing optimization using Digital Twins
2023, CEUR Workshop ProceedingsA Perspective on Stable Wired Networks while Reducing Energy Use with Load Balance
2022, Conference Proceedings - IEEE SOUTHEASTCONOn adaptive multi-objective optimization for greener wired networks
2020, SN Applied Sciences
- ☆
A preliminary version of this paper has been accepted for publication at the IEEE/IFIP Networking 2013 Conference. The conference paper version is provided as supporting material for this submission.