3130 IEICE TRANS. COMMUN., VOL.E95–B, NO.10 OCTOBER 2012 PAPER Router Power Reduction through Dynamic Performance Control Based on Traffic Predictions Hiroyuki ITO † , Nonmember, Hiroshi HASEGAWA † a) , Senior Member, and Ken-ichi SATO †b) , Fellow SUMMARY We investigate the possibility of reducing router power consumption through dynamic router performance control. The proposed algorithm employs a typical low pass filter and, therefore, is simple enough to implement in each related element in a router. Numerical experiments using several real Internet traffic data sets show the degree of reduction in power consumption that can be achieved by using the proposed dynamic performance control algorithm. Detailed analysis clarifies the relationships among various parameter values that include packet loss ratios and the de- gree of power savings. We also propose a simple method based on the leaky bucket model, which can instantaneously estimate the packet loss ratio. It is shown that this simple method yields a good approximation of the results obtained by exact packet-by-packet simulation. The simple method easily enables us to derive appropriate parameter values for the control algorithm for given traffic that may differ in different segments of the Internet. key words: router, performance control, energy efficiency 1. Introduction The amount of Internet traffic has rapidly and continuously increased all over the world [1], since broadband access such as ADSL and FTTH have become widely deployed. This trend will continue because broadband services such as 4-k cinema, ultra-high definition video [2] or e-science [3] are on the horizon. The rapid increase in traffic has resulted in huge power consumption levels at routers, which necessi- tates the development of energy efficient routers. Indeed, it is estimated that the power consumption of the telecommu- nication network will double from 2009 to 2017 due to the growth in the power consumption of Layer 3 IP/MPLS back- bones routers [4]. In order to reduce effectively the power consumption of future networks, minimizing the increase in power consumption in IP/MPLS core routers is essen- tial. Core routers consume an almost constant high level of power regardless of their loads. Indeed, Chabarek et al. [5] reported that router power consumption in the idle state is only 10% less than that for the fully loaded state. Almost the same data were reported in [6]. This is because Internet traffic rapidly fluctuates and so it is very difficult to predict future traffic volume, which is necessary to control routers. The processing performance of present routers is hence kept high so that they can process the maximum traffic volume at any time. A typical large core router consumes approxi- mately 30% of the total power in a packet processing engine Manuscript received January 17, 2012. Manuscript revised June 5, 2012. † The authors are with Nagoya University, Nagoya-shi, 464- 8603 Japan. a) E-mail: hasegawa@echo.nuee.nagoya-u.ac.jp b) E-mail: sato@nuee.nagoya-u.ac.jp DOI: 10.1587/transcom.E95.B.3130 and approximately 35% for its power supply and blowers [7]. The power consumption of the power supply and blow- ers increases in conjunction with the power consumption at the rest part of router. Thus, the router power consumption can be effectively reduced by reducing the power consump- tion of the processing engines, which consume almost the same level of power regardless of their processing load. In order to reduce the power consumption when the traffic volume is low, several studies considered dynamic performance control of transport systems to achieve energy savings [8], [9]. These studies investigated the effect of load concentration that is attained by traffic engineering, where traffic is concentrated at some routers, and the rest of the routers are turned off. A conventional study [8] proposed a method to achieve power saving in IP-WDM networks by turning off line cards or IP routers that are not necessary for each time interval. The information on traffic variation in a network must be given a priori and this minimizes the num- ber of network disruptions. In [9], Idzikowski et al. formu- lated an optimization for route assignment and line card ac- tivation to minimize energy consumption. The formulation uses all the traffic matrix information including future traffic volume for perfect planning and exact evaluation of power consumption. Although significant improvements were ver- ified, it is hard to apply this method directly to real networks due to the difficulty in obtaining exact predictions of future traffic. In [10], dynamic activation of links and route reloca- tion using only the observed traffic volume were considered. The candidate links to be turned off are selected preliminar- ily considering the connectivity between nodes. Other studies focused on dynamic performance control of the transport systems themselves whose processing abil- ity can be changed in a stepwise manner. The advantage of this approach is that the technique can be introduced locally without affecting other parts of a network and without apply- ing network wide control. Studies [11] and [12] focused on dynamically changing the link speed according to the cur- rent traffic information. In [11], the link speed of Ethernet ports of a PC and LAN are adaptively altered between two data rate states based on the current link or buffer memory utilization. This method is shown to be effective when traffic is bursty and the average utilization ratio is low. This occurs typically only when a few users are connected to the link. In [12], power consumption reduction in switches attained by employing link aggregation was proposed. Switches mea- sure the arriving traffic volume and manage the total capac- ity for each link, that is, increase or decrease in the num- Copyright c  2012 The Institute of Electronics, Information and Communication Engineers