A Sleeping and Offloading Optimization Scheme for Energy-Efficient ...

10 downloads 3495 Views 819KB Size Report
analysis, we investigate AP energy efficiency to obtain the sleep- awake threshold, which is used to select sleep or awake APs according to real-time status ...
IEEE COMMUNICATIONS LETTERS, VOL. 21, NO. 4, APRIL 2017

877

A Sleeping and Offloading Optimization Scheme for Energy-Efficient WLANs Chuan Xu, Member, IEEE, Zhenzhen Han, Guofeng Zhao, and Shui Yu, Senior Member, IEEE Abstract— In this letter, we propose an access point (AP) sleeping and user offloading optimization scheme to improve energy efficiency in densely deployed WLANs. Through real trace analysis, we investigate AP energy efficiency to obtain the sleepawake threshold, which is used to select sleep or awake APs according to real-time status information monitored on controller. Moreover, we formulate the user offloading problem as a reverse auction process to optimize energy efficiency of APs involved in offloading. Simulation results demonstrate that, comparing to traditional methods, our scheme can achieve up to 20% energy saving while maintaining effective system coverage and throughput. Index Terms— Energy efficient, sleep-awake, offloading optimization, WLAN.

I. I NTRODUCTION

F

OG computing extends Cloud Computing to edge of networks to provide customized applications and services for end users [1]. To support high connectivity and low latency fog services, a large number of wireless local area networks (WLANs) have been widely deployed in the edge of networks. In these WLANs, in order to supply enough capacity to meet the demands of peak hours, the density of APs is much higher than what normally needed. Meanwhile, during the offpeak period, as the capacity demand declining sharply, the utilizations of many APs are reduced to low or idle. However, the energy consumption of an idle AP is almost equivalent to a fully loaded one, which leads to a serious energy waste. To reduce this kind of wastage, many proposed solutions adopted switching strategies (also called sleep-awake) to turn off/on the low-utilization or idle APs to meet the current user needs [2]. The resource-on-demand strategy was firstly proposed for dense WLANs [3]. A green clustering algorithm was introduced to initiate a cycle of estimating user demand and performance to power on or off APs [4]. Furthermore, through real trace analysis, authors of work [5] proposed a simplified model to study AP switching frequency and energy saving, and presented a detailed investigation on AP turn off threshold and hysteresis window settings. However, due to user’s high mobility, the clustering methods depending on historical user behavior cannot adapt to network variation.

Manuscript received October 12, 2016; revised November 29, 2016; accepted December 14, 2016. Date of publication December 21, 2016; date of current version April 7, 2017. This work is supported by the National Science Foundation of China (NSFC) under grant 61402065 and Young Backbone Academics Funding Program of Chongqing Municipal Education Commission. The associate editor coordinating the review of this letter and approving it for publication was B. Rong. (Corresponding author: Guofeng Zhao.) C. Xu, Z. Han, and G. Zhao are with the School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, China (e-mail: [email protected]; [email protected]; [email protected]). S. Yu is with the School of Information Technology, Deakin University, Burwood, VIC 3125, Australia (e-mail: [email protected]). Digital Object Identifier 10.1109/LCOMM.2016.2642919

Moreover, an automatic sleep control strategy was proposed to execute the state changes of APs based on AP data traffic monitoring [6]. In [7], based on a centralized control framework, the actual network conditions in terms of both user density and traffic patterns were monitored and used to tune the energy consumption through a flexible energy-saving decision algorithm. Similarly, a context-aware power management framework and adaptive algorithms were proposed to dynamically configure different network elements according to user needs [8]. The authors of [9] proposed a cooperative energy management method to schedule wireless resources among gateways through monitoring, and implemented it within a federated WiFi network through a fully distributed protocol. Nevertheless, the energy consumption and user QoE were not considered in resource scheduling and user offloading of those methods. In this letter, aiming to solve the energy waste problem, we propose an AP sleeping coupled with user offloading optimization scheme to minimize energy consumption in WLANs. First of all, through real trace analysis, we inspect AP energy efficiency to pick the sleep or awake thresholds. Then, we select APs with low-utilization to sleep according to the realtime network status information monitored on the controller. Finally, we formulate the user offloading problem as a reverse auction to minimize the energy increasing from user offloading while maintaining effective throughput for users. Furthermore, we propose an energy saving scheme to reduce the energy consumption in WLANs. Simulation results demonstrate that when the number of users in WLANs is small, our scheme can reduce 80% energy consumption, and achieve up to 20% energy saving than clustering algorithm while maintaining effective system coverage and throughput. Moreover, when the number of users reaches up to half of system’s capacity, our scheme can reduce 50% energy consumption, which is much better than clustering algorithm and cooperative algorithm. II. P ROBLEM S TATEMENT AND M ODELLING A. Problem Statement A typical WLAN scenario with densely deployed APs is illustrated in Fig. 1. During the off-peak period, since there are only three users connected to A P3 and A P4, the utilization of these two APs are reduced to low, but they are consuming energy almost the same as if they were heavy loaded, which leads to serious energy waste. Therefore, if we can offload the small number of users on A P3 and A P4 to other APs, and turn off these two APs, the system energy can be saved. The effectiveness and potential benefit of AP offloading strategies strongly rely on two elements: 1) which APs to be turned off? (Sleep APs). The most important determinant of AP selection is AP’s energy efficiency. APs with lower energy efficiency should be turned off first. Meanwhile, the system coverage also

1558-2558 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

878

IEEE COMMUNICATIONS LETTERS, VOL. 21, NO. 4, APRIL 2017

TABLE I L IST OF PARAMETERS U SED IN THE S CHEME

Fig. 1.

The architecture of typical WLAN system.

needs to be sustained. 2) Offloading the users of the sleeping APs to which APs? The goal of user offloading is to reschedule the system resources to ensure the user QoE, and to minimize the increment of energy consumption. Benefiting from the extending of SDN and NFV technologies in heterogeneous wireless networks, resource scheduling for energy management in WLANs are becoming feasible. Firstly, the centralized architecture and the control plane and data plane decoupling supports real-time monitoring and flexible control of APs. Secondly, the abstraction of functionalities from the underlying physical infrastructure allows seamless handover for user offloading [10]. B. Modelling User Offloading As illustrated in Fig. 1, the running APs in WLAN are managed by the controller to provide access for the offloading users. Each A Pi has an unexploited capacity Ci of its bandwidth that it is willing to lease, unknown to the controller. When user j has the amount of traffic d j that needs to be offloaded, through cost evaluation of its current load, A Pi submits to the controller the bid [bi j , d j , Ci ], representing the price that it asks for offloading user j to itself. Then, the controller feedbacks the information to user j . User j pays A Pi the cost pi j to offload its traffic. To prevent market distortion, we force APs to provide the true information (bi j = v i j ). We define the utility of A Pi , u i , as the difference between the payment obtained from the users, pi j , and the valuation to offload user j to A Pi , v i j , according to  ui = (Pi j − v i j ). (1) j ∈Ui

III. E NERGY S AVING S CHEME In this section, we present the details of the AP sleeping coupled with user offloading optimization scheme. The parameters used in our scheme are described in Table I. A. Modelling AP Energy Efficiency for AP Selection To test the energy efficiency of AP devices, we conduct an experiment in a real network. We deploy 6 Netgear WNDR

Fig. 2.

Energy consumption and loss rate with throughput.

3800 AP devices in a common application scenario: working at 2.4 GHz with 802.11n mode, 20 MHz HT mode, random channel and 30dBm transmission power, and use power meter (TECMAN-TM6) to monitor AP energy consumption. The awake threshold (lmax ). The relationships between energy consumption, loss rate and throughput are shown in Fig. 2, when the throughput beyond 70Mbps, the loss rate suddenly breaks up from 0.94% and increases significantly, which leads to decreasing of user QoE. Consequently, we set the awake threshold to 70Mbps. To analyze the energy efficiency, we use least squares fitting regression analysis to construct a model to match the energy consumption and throughput. For improving the fitting accuracy, we increase the sampling rate to 160/s, and adopt piecewise fitting to obtain the energy consumption as follow  a1l 3 + b1l 2 + c1l + d1 , 0 ≤ l ≤ 40, (2) g = P(l) = a2l 3 + b2l 2 + c2l + d2 , 40 ≤ l ≤ 80, where l denotes throughput, g denotes energy consumption, and P(l) presents the energy consumption with l throughput. −1 We denote the energy efficiency by E(g) = d( Pdg(g)) . The sleep threshold (lmin ). The energy efficiency with throughput is shown in Fig. 3. There is a marginal cost effect

XU et al.: SLEEPING AND OFFLOADING OPTIMIZATION SCHEME FOR ENERGY-EFFICIENT WLANs

Fig. 3.

Algorithm 1 The Optimal Energy Saving Scheme Input: A U  lmin , lmax Output: Rs , Rw , x ij , pij 1. Monitor the load of APs; 2. Obtain Rs , Rw and A according to lmin , lmax , constraint (5) and (6); then, obtain U , A j , and Ui ; 3. Output dij , C i , ri ; 4. Obtain the x ij by solving MILP model; 5. Selecting and offloading users, ensuring pij according to (9); 6. According to Rs and Rw , sleep or wake up corresponding APs.

Energy efficiency with throughput.

in the energy efficiency, which reaches the maximum value with 20Mbps throughput. Obviously, 20Mbps throughput is an inflection point to the energy efficiency, and AP’s energy is under low-utilization when its throughput is smaller than 20Mbps. Consequently, we set the sleep threshold to 20Mbps. B. Optimal Auction for User Offloading We build a mixed-integer linear programming (MILP) model to obtain the optimal resource assignment, and adopt the second closed reverse auction mechanism to design the payment method to optimize the associated AP selection in user offloading. According to eq(1), the sellers use the reality valuation v i j as the bid,  l0 +di j v i j = P(di j ) = P  (l)dl, (3) l0

where l0 represents the load of A Pi before it offloads user j . Given the above definitions and notations, the reverse auction problem can be stated as follows. max S(x) = s.t.

 j ∈Ui



Rs 

[P(lk0 ) − P0 ] −

k=0

xi j

879



x i j ·v i j , (4)

i∈ A j ∈U

di j ≤ 1, ∀i ∈ A, ri j

(5)

x i j di j ≤ Ci , ∀i ∈ A,

(6)

x i j = 0, ∀i ∈ / A, ∀ j ∈ / U, x i j ∈ {0, 1} , ∀i ∈ A, ∀ j ∈ U.

(7)

j ∈Ui

(8)

Let S(x) denote the energy saved through the optimization of AP sleeping and user offloading. The first term of the objective function represents the reduced energy consumption through the sleep APs. lk0 is the load of A Pk before it offloads users. The second term represents the energy consumption increase on APs from offloading users to them. Constraint (5) is used to ensure the transmission rate requirement from user j to A Pi . Constrains (6) ensures that the load of A Pi after it offloads users will not exceed its capacity. Constrains (7) and (8) are binary decision variables. To improve the efficiency of auction, we adopt the second closed reverse auction mechanism to design the payment rule. According to the biding information submitted by buyers

and sellers, the controller arranges the bidding price of APs participating in the reverse auction for offloading user j in order, and selects the AP with lowest biding price as the winner,    mink=i bkj ,   Aj  > 1, pi j = (9)  A j  = 1. vi j , It is obviously that eq(9) satisfies the property of incentive compatibility and individual rationality. We prove the fairness as follows. Proposition 1 (Fairness): In order to get the maximum benefit, APs must give the bids according to their real valuation. Proof: A Pi can’t improve its own benefit without actually bidding in following cases. 1) When bi j > v i j , if pi j ≥ bi j , u i = pi j − v i j ≥ 0, the benefit of A Pi does not change; if pi j ≤ v i j , there is a bid lower than bi j , A Pi will lose and its benefit is zero; if v i j < pi j < bi j , there exist some bids lower than bi j , A Pi will lose and its benefit is zero. 2) When bi j < v i j , if pi j ≥ v i j , u i = pi j − v i j ≥ 0, the benefit of A Pi does not change; if pi j ≤ bi j , there is a bid lower than bi j , A Pi will lose and its benefit is zero; if bi j < pi j < v i j , u i < 0, the benefit is negative. Therefore, bi j = v i j is the best bidding strategy for APs. The implementation of the optimal energy saving scheme based on reverse auction is shown in Algorithm 1. The objective is to select the suitable APs for user offloading to minimize the energy consumption of WLAN. Based on the energy consumption function (2) and the energy consumption of sleep AP P0 in [9], the saved energy after using our energy saving scheme can be stated as follows, P = S(x) − (

Rw  h=0

P(lh0 ) −



P(∇dgh )),

(10)

g∈Rw

the first term represents the energy saved through the optimization of AP sleeping and user offloading, and the second term represents increased from awaking APs, where  l the energy  P(∇dgh ) = l00−dgh P (l)dl. IV. N UMERICAL R ESULTS In our simulation study, the density of AP is set to 0.001[AP/m2] as suggested in [5]. We deploy 100 APs and randomly place users in a 300m-by-300m area. The effective

880

Fig. 4.

IEEE COMMUNICATIONS LETTERS, VOL. 21, NO. 4, APRIL 2017

Comparison of energy saving.

our scheme has tiny differences with cooperative algorithm, but when more users connect to the system, our scheme can save nearly 5% energy than the cooperative algorithm. Comparing to the green-clustering algorithm, our scheme works much better, especially under low system load. Effective Coverage: As shown in Fig. 5, our scheme and cooperative algorithm provide better coverage for users than that of the green-clustering algorithm. The reason is that the users associated with APs are taken into account for sleep AP selection both in our scheme and the cooperative algorithm. The Impact on User Throughput: Fig. 6 shows that user’s average throughput provided by our scheme is stable around 3Mbps, which is 7% to 20% higher than that provided by the green-clustering algorithm and the cooperative algorithm. It demonstrates that our energy saving scheme will not cause any damage to user throughput when user offloading. V. C ONCLUSION

Fig. 5.

Comparison of coverage.

WLANs must be energy efficient to support fog computing environments. In this paper, we present an AP sleeping and user offloading scheme to reduce energy consumption. The result proves that our energy saving scheme can improve the energy efficiency significantly, while sustaining effective coverage and throughput for users. R EFERENCES

Fig. 6.

Average user throughput.

coverage radius of AP is set to 40m, the throughput of each user is set to a range from 2Mbps to 4Mbps, AP transmission power is set to 30dBm, and a classic coloring algorithm is applied to minimize the interference between the adjacent channels. In order to validate the effectiveness under different system loads, we change the number of users from 50 to 800 with a step length of 50. We conduct experiments to validate the efficiency of our energy saving scheme comparing with green-clustering algorithm [3] and cooperative energy-efficient method [9]. Energy Saving: As shown in Fig. 4, our proposed scheme can save 50% to 80% energy consumption under different system loads. When the system load is lower than 400 users,

[1] F. Bonomi, R. Milito, J. Zhu, and S. Addepalli, “Fog computing and its role in the Internet of Things,” in Proc. 1st Ed. MCC Workshop Mobile Cloud Comput., Aug. 2012, pp. 13–16. [2] L. Budzisz et al., “Dynamic resource provisioning for energy efficiency in wireless access networks: A survey and an outlook,” IEEE Commun. Surveys Tut., vol. 16, no. 4, pp. 2259–2285, 4th Quart., 2014. [3] A. P. Jardosh, G. Iannaccone, K. Papagiannaki, and B. Vinnakota, “Towards an energy-star wlan infrastructure,” in Proc. 8th IEEE Workshop Mobile Comput. Syst. Appl., HotMobile, Mar. 2007, pp. 85–90. [4] A. P. Jardosh, K. Papagiannaki, E. M. Belding, K. C. Almeroth, G. Iannaccone, and B. Vinnakota, “Green WLANs: On-demand WLAN infrastructures,” Mobile Netw. Appl., vol. 14, no. 6, pp. 798–814, 2009. [5] F. G. Debele, M. Meo, D. Renga, M. Ricca, and Y. Zhang, “Designing resource-on-demand strategies for dense WLANs,” IEEE J. Sel. Areas Commun., vol. 33, no. 12, pp. 2494–2509, Dec. 2015. [6] T. Tanaka, K. Abe, S. Aust, T. Ito, H. Yomo, and S. Sakata, “Automatic and cooperative sleep control strategies for power-saving in radio-ondemand WLANs,” in Proc. IEEE Green Technol. Conf., Apr. 2013, pp. 293–300. [7] K. Gomez, C. Sengul, N. Bayer, R. Riggio, T. Rasheed, and D. Miorandi, “MORFEO: Saving energy in wireless access infrastructures,” in Proc. IEEE 14th Int. Symp. Workshops World Wireless, Mobile Multimedia Netw. (WoWMoM), Jun. 2013, pp. 1–6. [8] N. Bayer, K. Gomez, C. Sengul, D. von Hugo, S. Göndör, and A. Uzun, “Load-adaptive networking for energy-efficient wireless access,” Comput. Commun., vol. 72, pp. 107–115, Dec. 2015. [9] C. Rossi, C. Casetti, C. F. Chiasserini, and C. Borgiattino, “Cooperative energy-efficient management of federated WiFi networks,” IEEE Trans. Mobile Comput., vol. 14, no. 11, pp. 2201–2215, Nov. 2015. [10] C. Xu, W. Jin, Y. Han, G. Zhao, and H. Tianfield, “MP-SDWN: A novel multipath-supported software defined wireless network architecture,” in Proc. 11th EAI Int. Conf. Commun. Netw. China, in press.