《[精品论文]Improving Energy-Efficiency of HFC Networks with a.doc》由会员分享，可在线阅读，更多相关《[精品论文]Improving Energy-Efficiency of HFC Networks with a.doc（11页珍藏版）》请在三一文库上搜索。
1、精品论文Improving Energy-Efficiency of HFC Networks with aMaster-Slave Linecard ConfigurationLu Ping, Yuan Yabo, Zhu Zuqing5(School of Information Science and Technology, University of Science and Technology of China, HeFei 230026)Abstract: We develop a novel traffic scheduling algorithm based on a mast
2、er-slave linecard (LC)configuration to improve the energy-efficiency of hybrid fiber-coaxial (HFC) networks. The algorithm forwards packets to the master or slave LC adaptively according to the traffic load, and toggles the LCs10between working and sleeping modes for energy-saving. To optimize the a
3、lgorithms key parameters, we model the system using a twodimensional Markov process and derive the analytical expressions of several performance metrics, including average packet delay, LC switching frequency, and energy efficiency improvement. We then verify the theoretical analysis with numerical
4、simulations using theMonte Carlo method. Both the theoretical and simulation results indicate that the proposed algorithm15can achieve significant energy efficiency improvement, while keeping the average packet delay and LCswitching frequency low.Keywords: Communication and information system; Hybri
5、d fiber-coaxial (HFC) networks; Energy-efficient scheduling; Master-slave linecard configuration; Energy-delay tradeoff200IntroductionNowadays, the increasing speed of equipment installations makes the Internets energy consumption a global concern. A recent study on network energy consumption growth
6、 indicated that access networks were among top carbon emission contributors in the Internet 1. Moreover, this situation will not change in the short-to-middle-term future, as the service providers are25investing heavily in broadband access networks to facilitate sufficient bandwidth for new network
7、applications, such as video-on-demand, tele-conferencing, video gaming and etc. To this end, many energy-saving techniques have been proposed for various types of access networks, including mobile access networks 2, passive optical networks (PON) 3, hybrid wireless-optical broadband access networks
8、(WOBAN) 4, digital subscriber line (DSL) networks 5, and hybrid30fiber-coaxial (HFC) networks 6, 7. HFC networks provide Internet data services over the existingcable television (TV) systems, which have an hybrid infrastructure that combines optical fibers and coaxial cables and deliver digitally mo
9、dulated signals through RF TV channels. Since this type of access networks have the second biggest user population globally among all wired ones 8,it is worth developing energy-saving techniques to reduce both the operational expenditure (OPEX)35and the environmental impacts.As the industry standard
10、 for developing HFC equipments, Data over Cable Service Interface Specifications (DOCSIS) 9 define two primary types of HFC network equipments, i.e., the cable modem terminal system (CMTS) located at an operators headend, and the cable modems (CMs) at customer premises. Recently, DOCSIS 3.0 10 has b
11、een released. It includes channel-bonding as40a major technology improvement, which can provide cable operators a remarkable opportunity to outpace competitors such as fiber-to-the-home (FTTH) providers. Specifically, DOCSIS 3.0 allows a CM to groom multiple RF channels from the CMTS as a virtual br
12、oadband channel for data transmission at 100 Mb/s or more. Even though it can boost up customers access data-rateseffectively, channel bonding also causes significant power consumption increase in HFC networks.Foundations: the Fundamental Research Funds for the Central Universities (WK2100060006); t
13、he NCET Project(NCET-11-0884); the SRFDP Project (20123402120014).Brief author introduction:Lu Ping, master student, research interests: queueing theory and green networks. Correspondance author: Zhu Zuqing, associate professor, research interests: green networks, optical networks, cloud computing a
14、nd multimedia networks. E-mail: zqzhuieee.org- 11 -45For instance, a CMTS linecard (LC) has to be designed with an increased number of ports, which makes it more power-hungry.Previously, we have reported several energy-saving algorithms for the HFC networks based on DOCSIS 3.0 6, 7. In 6, we conside
15、red the operations of a CMTS and CMs jointly and proposed an approach for cooperative energy-saving. In 7, we studied traffic scheduling in50channel-bonding CMs, and formulated an analytical model to investigate the energy-delay tradeoff.From the perspective of a CMTS chassis, multiple versions of L
16、Cs can co-exist. Although equipping the most advanced version of LCs in all slots of a CMTS chassis can boost its capacity to the highest, the capital expenditure (CAPEX) associated with the LC replacements and the OPEX caused by the energy bills can be prohibitively high. Recently, in 11, Parker et
17、 al.55proposed a master-slave router configuration for energy-saving in optical networks. A CMTSchassis can also incorporate a similar configuration and toggles between a master and a slave LCs for energy-saving. However, to make the master-slave configuration work efficiently in a CMTS, we still ne
18、ed to develop a traffic scheduling algorithm that considers the characteristics of HFC networks and to investigate the tradeoffs between energy efficiency and other performance metrics,60such as average packet delay and LC switching frequency.In this paper, we develop a novel energy-efficient traffi
19、c scheduling algorithm based on the master-slave LC configuration in a CMTS. The algorithm forwards packet to the master or slave LC adaptively according to the traffic load, and toggles the LCs between working and sleeping modes for energy-saving. To understand the performance impacts of the algori
20、thms key65parameters, we model the system using a two-dimensional Markov process and derive the analytical expressions of a few performance metrics, including average packet delay, LC switching frequency, and energy efficiency improvement. The theoretical analysis is then verified by numerical simul
21、ations using the Monte Carlo method. We also investigate the tradeoffs between the performance metrics to further optimize the algorithm.70The rest of the paper is organized as follows. Section 1 explains the system model and illustrates the operation principle of the proposed algorithm. In Section
22、2, we show the theoretical analysis. The performance evaluations are described in Section 3. Finally, Section 4 summarizes the paper.1System Model751.1 Second-order headline1.1.1Third-order headlineFig.1 shows the system model of the master-slave configuration of two LCs in a CMTS chassis. Notice th
23、at a CMTS chassis usually accommodates more than two LCs and the rest of the LCs can operate independently from these two. In the context of this work, we assume that the80upstream (US) capacities (i.e, the traffic capacity from CMs to a CMTS LC) of the master andslave LCs are identical, but the mas
24、ter LCs downstream (DS) capacity (i.e, the traffic capacity from a LC to CMs) is larger than that of the slave one. This assumption is valid since the traffic of HFC networks is asymmetric 12 and the DS provision speed of a CM is normally larger. Hence, we can trigger LC switching based on DS traffi
25、c only.85We assume that DS packets arrive at the CMTS following the Poisson process with a rate .The packets are then buffered in a finite queue. We denote the length of the queue as , in terms of number of packets. The number of pending packets in the queue is defined as for a time instant . We ass
26、ume the service time per packet on the two LCs follows the exponentialdistribution, where the mean values are denoted as and (), for the master90and slave LCs, respectively. The energy-efficient scheduling algorithm determines the operation modes of the LCs based on , and forwards packets to a prope
27、r LC. In a practical CMTS system, traffic switching between two LCs can be realized with the high-availability feature 13. The scheduling algorithm operates based on the definitions as follows.95Fig. 1 System model for master-slave linecard configuration100Definition 1 (LC Operation Modes). We defin
28、e two operation modes for a linecard: 1) Working, as it is turned on for transmitting data. The average power consumptions of the master and slave LCs in this mode are denoted as and (), respectively. 2) Sleeping, as it is in the energy-saving mode with an average power consumption of .We assume tha
29、t the power consumption of a CMTS LC increases sub-linearly with itscapacity14,105110115120125 (1)whereis a constant coefficient, andis the LCs capacity. Then, we get and .Definition 2 (Traffic Queue Sampling Period). The scheduling algorithm samples the trafficqueuewith a random period that follows
30、 an exponential distribution. The average value of is denoted as .Definition 3 (Linecard Switching Threshold). When a sampling period expires, thescheduling algorithm examines the length of the queue, . When , the master LC will be put into the working mode; otherwise, when , the algorithm will togg
31、le traffic transmission to the slave LC. Here,is a parameter that we use to optimize the algorithm for avoiding excessive LC switchings. To maximize energy-saving, we also assume that when the master LC becomes idle during the working mode ( ), it notices the scheduler to invoke a LC switching.Algor
32、ithm 1 illustrates the details of the proposed scheduling algorithm. We assume that theswitching transition is negligible since it is usually much smaller than in a practical CMTSsystem.Algorithm 1 Energy-Efficient Scheduling for Master-Slave Configuration1: while system is operational do2: measure
33、L(t);3: if L(t) s + then4: if slave LC is in working mode then5: invoke a LC switch: slave _ master;6: put slave LC into sleeping mode;7: end if1301351401458: else if L(t) s then9: if master LC is in working mode then10: invoke a LC switch: master _ slave;11: put master LC into sleeping mode;12: end
34、 if13: end if14: determine the next queue sampling period Ts;15: wait (Ts);16: end while2Theoretical Analysis2.1 Infinitesimal Generator Matrix We modelthesystemdiscussedabovewithatwo-dimensionalMarkov process , where is the pending packets in the queue, and identifiestheLCsstate,with and representt
35、hestateof and, respectively. For the Markov process, we define the subset of the states that have a same as a level, denoted as . Hence, the state space can be written as, whereWe define the traffic load as the ratio betweenand, i.e., . The infinitesimal generator matrix of the Markov process can be
36、 written as:(2)150where the sub-matrixdenotes the backward transition rates from level to , sub-matrix is for the local transition rates within level , and sub-matrix is for the forward transition rates from level to . These sub-matrices can be derived as follows with ,(3)(4)(5)1551601652.2 Matrix G
37、eometric SolutionWe define the steady probabilities of the Markov process as, (6)withas the row-vector of the steady probabilities of level ,It is known that the Markov process can satisfy 15, (7)and the row-vectors should make sure that the summation of the steady probabilities equals 1,(8)170175wh
38、ereis a column-vector of 1s. Then, by combining Eqns. (refeq:qbd-refeq:qbd2), can be analytically expressed with, and .2.3 Performance MetricsIn this Subsection, we provide the analytical expressions of several system performance metrics, including average packet delay, LC switching frequency, and e
39、nergy efficiency improvement based on the above theoretical analysis.2.3.1Average Packet DelayAccording to the Littles Law, the average packet delay can be obtained as (9)2.3.2Linecard Switching FrequencyThe LC switching frequencycan be derived as,2.3.3Energy Efficiency ImprovementThe energy-efficie
40、nt improvement, can be calculated as,(10)180(11)1853Performance Evaluations and Algorithm OptimizationsWe perform Monte Carlo simulations with 1000 time instants to verify the theoretical analysis discussed above and to evaluate the impacts of the algorithms parameters, including the capacity ratio
41、between the master and slave LCs , the average traffic sampling period, and the switching threshold parameter. The simulation parameters are listed in the Tab. 1.Tab. 1 Simulation ParametersN, System queue length 1000 packets m, Master LCs service rate per time unit 100 packets s, Slave LCs service
42、rate per time unit 1-100 packets Pm, Master LCs average power in working mode 1 power-unitP0, A LCs power in sleeping mode 0 power-unit, Traffic load 0-11901952003.1 Capacity ratio between master and slave LCsIn this Subsection, we fix,to investigate the impacts of on , , and, defined in Eqns. (refe
43、q:delay-refeq:efficiency). Figs. 2-4 plot the results from both theoretical calculations and numerical simulations. It can be seen that the analytical results match well with the numerical ones. Fig. 2 shows the impact of on average packet delay . The results indicate that has a sudden decrease when
44、. This is because that when , the slave LCs capacityis catching up with the traffic load . The impact ofon LC switching frequency shown in Fig. 3 indicates that decreases withand becomes 0 when . Fig. 4 illustrates the results on energy efficiency improvement . We observe that when This can be expla
45、ined as that when, the master-slave configuration becomes a master-master configuration that is equivalent to the normal operation without any energy-saving. In the other extreme case when , the system reduces to a single LC configuration with the sleeping-mode based energy-saving. Hence, when ,approaches to its maximum value . Notice that also results i