## Journal of Green Engineering

Vol: 4    Issue: 1

Published In:   October 2013

### Energy-Efficient Traffic Engineeringfor Future Networking Infrastructures

Article No: 1    Page: 1-12    doi: https://doi.org/10.13052/jge1904-4720.411

 1 2 3 4 5

Energy-Efficient Traffic Engineering for Future Networking Infrastructures

Received: December 2, 2013; Accepted: February 20, 2014: Publication: May 14, 2014

George Athanasiou

• KTH Royal Institute of Technology, 100–44, Stockholm, Sweden Email: georgioa@kth.se

## Abstract

A general problem formulation for energy-efficient traffic engineering for future core networks is presented. Moreover, a distributed heuristic algorithm that provides jointly load balancing and energy efficiency is proposed. Simulation results show that the proposed algorithm approaches the optimal network operation, in terms of throughput and energy consumption.

## Keywords

• Network Management
• Traffic Engineering
• Energy-awareness

## 1 Introduction

While the networked communication systems are continually evolving, network and service management is crucial to ensure proper operation (as far as configuration, performance, faults, security issues and accounting are concerned). Nowadays, expert human resources and complex systems are required to manage the increasing plethora of networked devices, ranging from small sensors to terabit routers, and the large variety of applications. The explosion of the Internet and the proliferation of networked devices create unique challenges for network and system management. Moreover, the complexity of networked systems and the cost of management are also constantly growing. Therefore, novel traffic engineering solutions are needed in order to guarantee the efficient operation of the next generation networks.

Traffic Engineering (TE) plays a crucial role in determining the performance and reliability of network deployments. A major challenge in traffic engineering is how to cope with dynamic and unpredictable changes in traffic demands and how the network could handle possible traffic variations in a way that load balancing, congestion avoidance and efficient service provisioning are ensured. In this direction, TE approaches must apply efficient resource optimization strategies so as to eliminate these effects. A more straightforward explanation of TE is given in [1]: “to put the traffic where the network bandwidth is available”. Therefore, the nature of TE is effectively a kind of routing optimization for enhancing network service capability to achieve the aforementioned objectives.

Recently, there is an increasing interest in providing energy-aware network operation. The rapid growing of the users and the services that must be supported, the spreading of broadband access in conjunction with the increased energy prices affected the demand for energy-aware service provisioning. Unfortunately, the current underlying network infrastructures, namely routers, switches and other network devices, lack effective energy management solutions.

In this paper we provide a joint problem formulation for optimal energy-aware load balancing in the network. Then, we propose a distributed ENergy-Aware TRaffic Engineering (ENTRE) scheme that smoothly introduces the aforementioned major issues in real network deployments.

## 2 Related Work

Traffic Engineering is a widely studied topic in literature. Fortz et al. were the first to propose the idea of IGP link weight optimization [2], [3]. Several other representative approaches could be classified into the following categories: Intradomain and Interdomain [4], MPLS-based and IP-based [5], [6], Offline and Online [7], [8], Unicast and Multicast [9], [10].

A challenging task is to identify the main parts of the Internet that dominate its power consumption and investigate methods for improving energy consumption [11]. The first attempt to introduce energy savings in the Internet was made in [10]. Moreover, several approaches studied the problem of managing energy consumption in end user devices [12] and LANs [13], [14].

Recently, the authors in [15] discuss the idea of dynamically turning part of the network operations into sleeping mode, during light utilization periods, in order to minimize the energy consumption. The authors in [16] identified the power saving problem in the Internet, and propose sleeping as the approach to conserve energy. In their approach they support uncoordinated sleeping which works at link level and coordinated sleeping which operates at network level. Moreover, routing, rate adaptation and network control are mobilized towards energy-efficient network operation [1719], [26], [27].

Chabarek at al. [20] introduced power awareness in network design and routing and they conduct valuable experiments with popular routers and create a router power consumption model.

Lastly the authors in [21] propose an intra-domain traffic engineering mechanism, which maximizes the number of links that can be put into sleep under given performance constraints such as link utilization and packet delay.

Unfortunately, none of these approaches provide problem formulation in the direction of jointly studying the “traditional” objectives and the new objectives (energy-awareness) of TE. This paper is an attempt to put these objectives under a joint problem formulation and to propose lightweight solutions that could be applied in real network deployments.

## 3 Traffic Engineering Approach

In this section we give a general formulation of the joint load balancing and the energy efficiency problems. Then, we present a distributed ENergy-Aware TRaffic Engineering (ENTRE) scheme that follows the guidelines provided by analysis of the joint problem formulation.

### 3.1 Network Model

Figure 2.1 depicts our network model, where each ingress router may have traffic demands for a particular egress router or set of routers. Multiple paths (MPLS tunnels) are used to deliver traffic from the ingress to the egress routers. Traffic is split among the available paths at the granularity of a flow, to avoid reordering TCP packets or similar effects that lead to performance degradation (using efficient traffic splitting approaches, like [22]). Moreover, we consider that the paths are computed and re-computed (if it is necessary) offline by the operator, since most of the operator's networks work in this way.

Figure 1 Network topology

Formally, we assume that for each ingress-egress node pair i the traffic demand is Ti and multiple paths Pi could be used to deliver the traffic from the ingress to the egress node. Therefore, a fraction of the traffic xip is routed along path p(p ∈ Pi). In addition, the energy consumption of an active link is affected by the maximum rate that the link can support (10Mbps, 100Mbps, etc) and the current utilization of that link. The calculation of the energy consumption of link l, el, with capacity cl, is based on the simple model proposed in [23] (used also in several approaches in literature):

$el=PowerConsumption(cl)×UtilizationFactor(l)$

PowerConsuption(cl) is the base power consumption of a link with capacity cl and UtilizationFactor(l) is the scaling factor to account for the utilization of link l. Table I contains the definition of the variables used in our problem formulation.

Table 1 Variables

Variables Description
L Set of links in the network
IE Set of Ingress to Egress node pairs
el Energy consumption of the port connected to link l
Pi Set of paths of Ingress to Egress node pair i
Ti Traffic demand of Ingress to Egress node pair i
xip Fraction of traffic of Ingress to Egress node pair i, sent through the path p
rip Traffic of Ingress to Egress node pair i, sent through path p
Pl Set of paths that go through link l
Li Set of links that are crossed by the set of paths Pi
${B}_{p{t}_{m}}$ Number of bits sent along the path p during tm seconds

### 3.2 Joint Problem Formulation

In our formulation, the optimal splitting of the traffic of each ingress-egress node pair along the available paths is performed with the objective: Assure that the maximum link utilization (total traffic on an active link divided by the link capacity) in the network is minimized. In this way resource-efficient (in this case link utilization/bandwidth) and balanced/stable network operation is achieved [24] ).

However, in the previous policy there are no guarantees related to the energy consumption in the network. In order to introduce energy-awareness we raise a second objective: “Assure that the energy consumption of the active routes is balanced in the network. That is, find the route with maximum energy consumption and minimize it. In this way, resource-efficient (in this case energy consumption) and balanced/stable network operation is achieved [24]

In our try to avoid “Resource Gluttony” (in terms of bandwidth and energy) we combine the previous single objectives in a unified objective function that takes into account the link utilization and the energy consumption of the route(s) that the link belong(s) to:

$minxipmaxl∈Lp∑i∈IE∑p∈Pi(xipTicl×∑k∈Lpek)​​, subject to: xip≥0,∀p∈Pi,∀i∈IE cl≥∑i∈IE∑p∈PlxipTi,∀l∈L ∑p∈Pixip=1,∀i∈IE xip=[0,1],∀p∈Pi,∀i∈IE$

The previous constraints ensure that: the fraction of traffic for a specific ingress-egress node pair sent across a path cannot be negative, the capacity of each link cannot be outreached and the traffic splitting through the available paths meets the traffic demands.

### 3.3 Energy-Aware Traffic Engineering (ENTRE) Heuristic

We now present an ENergy-Aware TRaffic Engineering (ENTRE) heuristic that follows the previous model and applies an online distributed Traffic Engineering approach that jointly balances load and energy consumption in realtime, responding to actual traffic demands. ENTRE uses multiple paths to deliver traffic from an ingress to an egress node, moving traffic from over-utilized to under-utilized paths. In these adaptive actions we take into account the energy consumption of the multiple routes. Consequently, the main contribution of our approach is to provide dynamic and lightweight management of the load and the energy consumption, avoiding in this way “resource gluttony” in the network.

In our algorithm, each ingress-egress node pair i measures every tm seconds a change in the fraction of traffic (Δxip) sent along path p. Furthermore, ENTRE measures the energy “distance” (ΔEip) of path p from the average energy consumption of the paths between ingress-egress node pair i:

$Δxip=(U¯i−Uip)rip∑k∈Pirik,whenUip>UminΔEip=(E¯i−Eip),whenEip>Emin$

where:

$rip=Bptmtm,∀p∈Pi,∀i∈IEU¯i=∑p∈PiripUip∑k∈Pirik,∀p∈Pi,∀i∈IEE¯i=∑p∈PiripEip∑k∈Pirik,∀p∈Pi,∀i∈IE$

In case that Δxip > 0 (p is underutilized), the fraction of traffic sent along path p must be increased by Δxip. Contrary, in case that Δxip < 0 (p is over-utilized), the fraction of traffic sent along path p must be decreased by Δxip. It is obvious that a possible increase in the fraction of traffic sent along a path will lead to energy consumption increase (the increased maximum utilization of that path will affect the energy consumption). Since one of the main objectives of our energy-aware approach is to keep also the maximum energy consumption in low levels, we apply the same policy for the energy consumption of the paths (ΔEip > 0: Energy consumption could be increased in p, ΔEip < 0: Energy consumption must be decreased in p). ENTRE combines the previous policies, balances the traffic every tm seconds and jointly keeps the maximum link utilization and the path energy consumption as low as possible. We summarize the basic rules in our heuristic mechanism:

1. IF: Δxip > 0 and ΔEip > 0 DO: Apply Δxip to p
2. IF: Δxip < 0 and ΔEip < 0 DO: Apply Δxip to p
3. IF: Δxip > 0 and ΔEip < 0
1. IF: ${\overline{E}}_{i}-{E}_{ip}>{T}_{E}$ DO: Exclude p from the routing table and turn the corresponding links into sleeping mode. Traffic is proportionally provisioned to the remaining paths.
2. ELSE DO: Nothing
4. IF: Δxip < 0 and ΔEip > 0
1. IF: ${\overline{U}}_{i}-{U}_{ip}>{T}_{U}$DO: Apply Δxip to p
2. ELSE DO: Nothing

It is true that paths with higher minimum capacity need more traffic to achieve the same utilization as smaller capacity paths. This is the main reason why Δxip is normalized by the rates. This makes the change in a path's traffic proportional to its current traffic share. Lastly, we would like to note that in the execution of ENTRE the constraints described in the previous subsection (in the joint problem formulation) must be respected.

### 3.4 Implementation of ENTRE in Real Network Infrastructures

In order to give a thorough presentation of the proposed mechanism we discuss several deployment issues that arise when trying to apply the ENTRE scheme in real network deployments. Firstly, ENTRE is executed at each IE node pair in the network and the main decision mechanism is executed at the ingress nodes. Moreover, supposing that we support MPLS-based operation, we require several LSPs for each IE node pair in order to split the traffic to the available routes (the current ISP-class routers can support up to 16 LSPs). Traffic splitting is performed seamlessly using sophisticated mechanisms [14]. In addition, for each IE node pair MPLS-based monitoring is performed (probe request/response) in order to estimate the network and flow performance. Lastly, ENTRE is implemented on top of the router functionality (software package) handling the basic functionalities that are offered (sleeping mode, etc.).

## 4 Evaluation

We present the evaluation study of the proposed heuristic scheme compared to the optimal solutions. We consider a network topology where four ingress nodes send traffic to four egress nodes (20 routers in total). We are using OMNET++ to simulate ENTRE and IBM ILOGCPLEX Optimizer to find the optimal solutions. Figure 2.2a depicts the network throughput while the number of disjoint paths grows and Figure 2.2b depicts the energy consumption while the achieved network throughput grows (ENTRE is compared to OSPF [[1]]). Lastly, Table 2 presents the performance of ENTRE, in different scenarios, compared to the optimal energy saving in the network.

Figure 2 Network throughput and energy consumption. a Throughput (OSPF vs. our approach) and b Energy consumption (OSPF vs. our approach).

Table 2 Entre performace

## 5 Conclusions

The joint modeling of balanced and energy-efficient network operation inspired the design of a heuristic approach that tries to meet the requirements of the future core networks. The simulation results show that the proposed approach tends to behave like an optimal load balancer in the network, influenced by the minimization of the energy consumption. ENTRE converges after a small number of iterations, proving in this way it's lightweight operation.

## References

[1] Y. Lee et al., “Traffic Engineering in Next-Generation Optical Networks,” IEEE Commun. Surveys & Tutorials, vol. 6, no. 3, 2004, pp. 16–33.

[2] B. Fortz and M. Thorup, “Internet Traffic Engineering by Optimizing OSPF Weights,” in Proceedings of IEEE INFOCOM, 2000.

[3] B. Fortz, M. Thorup. Optimizing OSPF Weights in a Changing World. IEEE JSAC, 2002.

[4] R. Teixeira, T. Griffin, A. Shaikh, and G.M. Voelker, Network Sensitivity to Hot-Potato Disruptions, ACM SIGCOMM, August 2004.

[5] Awduche, D.O., MPLS and Traffic Engineering in IP Networks, IEEE Commun. Mag., vol. 37, no. 12, Dec. 1999.

[6] B. Fortz, J. Rexford, M. Thorup, Traffic Engineering with Traditional IP Routing Protocols, IEEE Commun. Mag., vol. 40, no. 10, Oct. 2002.

[7] D. K. Goldenberg, L. Qiu, H. Xie, Y. R. Yang, and Y. Zhang, Optimizing Cost and Performance for Multihoming, ACM SIGCOMM 2004.

[8] A. Elwalid, C. Jin, S. Low, I. Widjaja, MATE: MPLS Adaptive Traffic Engineering, IEEE INFOCOM, 2001.

[9] M. Kodialam, T.V. Lakshman, Minimum Interference Routing of Applications to MPLS Traffic Engineering, IEEE INFOCOM, 2000.

[10] M. Kodialam, T.V. Lakshman, S. Sengupta, Online Multicast Routing with Bandwidth Guarantees: A New Approach Using Multicast Network Flow, IEEE/ACM Trans. Networking, vol. 11, no. 4, Aug. 2003.

[11] K. Hinton, J. Baliga, M. Feng, R. Ayre, R.S. Tucker, Power consumption and energy efficiency in the internet, IEEE Network Magazine, Special Issue on “Energy-Efficient Networks,” vol. 25, no. 2, pp.6–12, March-April 2011.

[12] Y. Agarwal, S. Hodges, J. Scott, R. Chandra, P. Bahl, and R. Gupta. Augmenting Network Interfaces to Reduce PC Energy Usage. In NSDI, 2009.

[13] C. Gunaratne, K. Christensen, and B. Nordman. Managing Energy Consumption Costs in Desktop PCs and LAN Switches with Proxying, Split TCP Connections, and Scaling of Link Speed. Int. J. Netw. Manag., 15(5): 297–310, 2005.

[14] M. Gupta, S. Grover, and S. Singh. A Feasibility Study for Power Management in LAN Switches. In ICNP, 2004.

[15] R. Bolla, R. Bruschi, A. Cianfrani, M. Listanti, Enabling Backbone Networks to Sleep, IEEE Network Magazine, Special Issue on “Energy-Efficient Networks,” vol. 25, no. 2, pp. 26–31, March–April 2011.

[16] M. Gupta and S. Singh. Greening of the Internet. In SIGCOMM, 2003.

[17] A. Cianfrani, V. Eramo, M. Listanti, M. Marazza, E. Vittorini, An Energy Saving Routing Algorithm for a Green OSPF Protocol, IEEE INFOCOM 2010, San Diego (USA), 15 March – 19 March 2010.

[18] S. Nedevschi, L. Popa, G. Iannaccone, S. Ratnasamy, D. Wetherall, Reducing Network Energy Consumption via Sleeping and Rate-Adaptation, ACM, USENIX, NSDI, 2008.

[19] N. Vasic, D. Kostic, Energy-Aware Traffic Engineering, in e-Energy '10 Proceedings of the 1st International Conference on Energy-Efficient Computing and Networking

[20] J. Chabarek, J. Sommers, P. Barford, C. Estan, D. Tsiang, and S. Wright. Power Awareness in Network Design and Routing. In INFOCOM, 2008.

[21] M. Zhang, C. Yi, B. Liu, B. Zhang, ”GreenTE: Power-Aware Traffic Engineering”, IEEE ICNP 2010, Kyoto, Japan.

[22] S. Kandula, D. Katabi, S. Sinha, A. Berger, Flare: Responsive Load Balancing Without Packet Reordering, ACM Computer Communications Review, 2007.

[23] P. Mahadevan, P. Sharma, S. Banerjee, and P. Ranganathan, “A Power Benchmarking Framework for Network Devices,” in Proceedings of IFIP Networking, May 2009.

[24] D. Bertsekas, R. Gallager, Data Networks. Englewood Cliffs, NJ: Prentice-Hall, 1992.

[25] S. Krasser, H. Owen, J. Sokol, J. Grimminger and H-P. Huth, “Adaptive measurement-based traffic engineering in small differentiated services domains”, EUROPEAN TRANSACTIONS ON TELECOMMUNICATIONS, Volume 18, Issue 1, January/February 2007

[26] G. Athanasiou, K. Tsagkaris, P. Vlacheas, D. Karvounas and P. Demestichas, “Multi-Objective Traffic Engineering for Future Networks”, in IEEE Communications Letters Journal, Volume: 16, Issue: 1, 2012, ISSN: 1089-7798

[27] K. Tsagkaris, G. Athanasiou, M. Logothetis, Y. Kritikou, D. Karvounas and P. Demestichas, “Introducing Energy Awareness in the Cognitive Management of Future Networks”, in Journal of Green Engineering, River Publishers, Vol: 1, Issue: 4, ISSN: 1904–4720, July 2011

## Biography

George Athanasiou received the diploma in Electrical and Computer Engineering from University of Thessaly, Greece in 2005. He obtained his M.Sc. and Ph.D. degrees in Electrical and Computer Engineering from the same University, in 2007 and 2010 respectively. Currently, he is a research scientist at KTH Royal Institute of Technology, Electrical Engineering School and ACCESS Linnaeus Center, Department of Automatic Control, Stockholm, Sweden. He is also co-founder and CIO of Aukoti AB, a Swedish startup on sensor networking and building automation. His research interests include the design and performance evaluation of wireless networks, resource management, service and network management, cognitive networking and optimization techniques. He has authored numerous publications in these areas in international journals and refereed conferences. He is a member of the IEEE and ACM and the Technical Chamber in Greece.