Algorithms, Design, Performance
Keywords
IEEE 802.16-2009; power saving class (PSC); energy saving;
multiple MSSs; state transitions
1. INTRODUCTION
IEEE 802.16 Worldwide Interoperability for Microwave Access (WiMAX) is used for the internet of Broadband Wireless Access (BWA), and its characteristics of widespread coverage area in metropolitan areas and high-speed bandwidth are much better than those of personal communication networks we use today. By adopting the structure of Point-to-Multipoint (PMP), a Base Station (BS) can serve several Mobile Subscriber Stations (MSSs) with multiple connections simultaneously, but the scheduling algorithm can not violate each Quality of Service (QoS) demand of connections.
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.
"IWCMC’10, June 28– July 2, 2010, Caen, France. Copyright © 2010 ACM 978-1-4503-0062-9/10/06/...$5.00"
Because the scheduling algorithm is an open issue on the Media Access Control (MAC) layer in the IEEE 802.16-2009 standard (including the specification of IEEE 802.16e) [1], an efficient scheduling algorithm must increase the capacity of BS in order to serve more MSSs in large-scale environments. Besides, the algorithm attempts to make MSSs work longer based on the battery-powered energy. In the standard, there are four QoS mechanisms: Unsolicited Grant Scheme (UGS), real-time Polling Service (rtPS), non-real-time Polling Service (nrtPS) and Best Effort service (BE) [2]. UGS, like VoIP, requests consistent bandwidth for a guaranteed period of time. Unlike UGS, rtPS, like streaming audio or video, requests variable bandwidth and strict delay bound. nrtPS, like FTP service, provides efficient service of non-real-time traffic with minimum reserved rate. Like web browsing, BE allocates bandwidth only when the bandwidth is not exhausted. In this paper, we focus on UGS applied to multiple connections among multiple MSSs
The standard also establishes three types of Power Saving Class (PSC): type I, type II and type III. Even though the process of each type is different, the goal is identical to less energy consumption for each MSS. In the PSC, there are two states of transceiver within the time unit of frame: on-state and off-state.
The continuous frames with on-state are called listening period, and those with off-state are called sleep period. We can find that an association between listening period and sleep period is a period has no packet to deliver. When applying to UGS and Real-Time Variable Rate (RT-VR), type II will repeat sleep cycle sequentially; i.e. the sleep cycle has a fixed listening period and sleep period. When the purpose of multicast or management operations is known beforehand, type III only enters the sleep period once. Fig. 1 presents a comparison between these types [3].
Because we focus on the UGS of QoS type, we adopt the Type II of PSC in this paper.
In the real world, a MSS usually has multiple connections simultaneously; furthermore, a BS also serves many MSSs at the same time. The problem is that the BS can only deliver packets to a MSS in a time slot, the unit of minimal time for scheduling.
When PSC is adopted, the whole frame, the unit of minimal time for PSC, must awake if a slot wants to deliver a packet. In past research, many algorithms had been proposed to reduce the consumption of power, but most of them only considered multiple connections in a MSS, which does not fit in with real world
situations. On the contrary, others proposed algorithms considering the situation of multiple MSSs with multiple connections; nevertheless, it is difficult to increase the amount of MSSs. In this paper, we propose an efficient algorithm, which refers to both categories and avoids state transitions. The result shows that our scheduling algorithm can serve more multiple MSSs with multiple connections and still maintain high sleep ratio for energy efficiency.
Figure 1. Power saving classes defined in the IEEE 802.16-2009 The remainder of this paper is organized as follows. In section II, we not only review the related work but also classify it into two categories; furthermore, we summarize the problems. In section III, we propose a new scheduling algorithm to match the real world situations. The simulation results are presented in section IV, and in conclusion, we summarize this paper in section V.
2. MOTIVATION AND PROBLEM DEFINITION
We categorize the related work into two categories: the first category concentrates on power saving with multiple connections in only a single MSS, and the second category deals with multiple connections among multiple MSSs. There are some different properties between these two categories, particularly in combining or separating the overlapping frames. In the related work, we find that [4-6] and [10] focused on the first category and [7-9] and [3]
emphasized the second one. Both categories consider three basic parameters. The first parameter is packet inter-arrival time representing the interval period, in which two packets are delivered between BS and MSS. In type II of PSC, inter-arrival time must be the same in the sleep cycle repetitively. The second parameter is bandwidth, which means the maximal traffic that BS grants for MSS in a frame. If the number of MSS increases, the bandwidth for each MSS decreases. The third parameter is delay bound, which is used to concern the maximum tolerable period of inter-arrival time. In this paper, we only consider the scenario that inter-arrival time is smaller than delay bound. In fact, it is almost true in the real world.
References [4-5] propose the Maximum Unavailability Interval (MUI), and apply the Chinese Remainder Theorem to select the best start frame for each connection. The purpose of increasing the overlapping number of frames in the sleep period among connections is to decrease the energy consumption of MSS. In addition, [5] proposes Intelligent Table Consulting (ITC) to reduce the computational complexity, but it does not consider the bandwidth and how sleep cycle among numerous connections is scheduled. While [5] only changes the start frame on each connection, the result represents to improve the limited consumption of power. All of [3-10] consider delay bound on scheduling algorithms with a single PSC; more importantly, [6]
attempts to consider with multiple PSCs. When the size of packet is smaller than bandwidth, [6] derives the most efficient power saving rate among all other papers; nonetheless, when scheduling, the algorithm of [6] does not consider the state transitions, a phenomenon proposed in [10]. In fact, [10] is the first paper to discuss this new parameter that inherited from [11] to improve Aperiodic on-off Scheme (AS) [3] and Minimum Wakeup Time (MWT) [7] in 802.16. Although the scheduling of [10] can work more efficiently than [3] and [7], sleep cycle changes from periodic to aperiodic. In other words, the scheduling of [10] still causes the issue of state transitions. We consider this new parameter in our algorithm and try to avoid it.
Reference [7] proposes an algorithm separating different transmission time among MSSs, but it is deficient. Reference [8]
fixes this problem by adopting Ford-Fulkerson algorithm.
Although [8] can work more efficiently than [7], it has the same problem as [10]. Reference [9] proposes a much easier and more efficient algorithm to interleave different MSSs and still maintains each connection with periodic sleep cycle; in other words, each connection among MSSs can still fit in with UGS. Unfortunately, this algorithm brings the unnecessary cost because of additional connection to MSS and does not discuss power saving. Besides, [9]
also assumes that each MSS only has one connection, but it is not real world situation. Reference [3] proposes Periodic on-off scheme (PS) and AS and considers to combine not only the packets from different connections but also the pattern from different MSSs under the maximum bandwidth of minimum delay bound. Before [6] and [10] are proposed, [3] is more outstanding in power saving than previous research; nonetheless, [6] and [10]
do not consider the situation of multiple MSSs that [3] does. In the simulation result, we compare PS with a new scheduling algorithm called Power Saving Class Management Scheme based on CAGE (PSS-CAGE), which we propose in this paper, with PS.
In Fig. 2, we can observe that PS has a limited BS pattern to the minimal MSS pattern; on the contrary, PSS-CAGE has a limited BS pattern to the maximal MSS pattern. For this reason, we can estimate that the number of MSSs, which BS can maintain, increases when the length of MSS patterns is much more variable.
For example, the BS pattern in Fig. 2(b) can serve an additional MSS pattern 1 and an additional MSS pattern 2, but that in Fig.
2(a) can only serve an additional MSS pattern 1.
Figure 2. Example of difference between (a) PS and (b) PSS-CAGE
In this paper, we want to combine the issues of two categories.
We define that Bf is the maximum bandwidth per frame granted from BS in an OFDM frame for a MSS. We attempt to aggregate potential frames from each connection as more as possible, and still keep each delay bound of connections, as shown in Fig. 3(a), and the maximum bandwidth of MSS pattern under control. In different MSSs, we interleave the overlapping frame by rearranging the start frame of each type II connection but not violate the delay bound. By the research of [9], we can find that if
the sleep cycles of two connections are multiples or factors of each other, the sleep cycles do not have overlap with each other, and they can work together in the same MSS. The same property can apply to two MSS patterns in the same BS. Therefore, we must maintain each MSS pattern to conform to type II in order to interleave with other MSSs as shown in Fig. 3(b).
Figure 3. Example of ideas
Because the probability that sleep cycle of connections or MSS pattern becomes the multiple or factor of each other will increase, the bandwidth utilization rate of BS pattern will increase, when the number of connection and MSS increases. On the other hand, in some special cases such as shown in Fig. 4(a), MSS pattern is not a multiple or factor of each other but can still interleave successfully. Nonetheless, in some special cases such as shown in Fig. 4(b), MSS patterns will overlap in some frames. This means that those MSSs can not work simultaneously in the same BS. In order to simplify the situation of Fig. 4(a) and avoid the situation of Fig. 4(b), we can use the relation between multiple and factor to schedule the BS patterns like in Fig. 3(b). Furthermore, we must schedule the connections of each MSS and consider all four parameters, especially state transitions, referred in the previous paragraph.
Figure 4. Example of special case
3. PROPOSED SCHEMES
We define that Ti is the length of MSS i, Ts is the minimal length Algorithm 1: Scheduling BS pattern
Input:
Sorting Ti from minimum to maximum;
Ts = T1; Tm = T2; Mu = Tm / Ts; S1 = (T1*Bf) –TFR1; CHECK=1;
Join MSS pattern 1 to BS pattern;
Output:
Checking whether Ti can join to the BS pattern For i = 2 to a do we can avoid the problem presented in Fig. 4(a) and Fig. 4(b).
Figure 5. Example of Algorithm 1
The remaining work is how to schedule each connection to a sleep cycle of type II in the same MSS and maintain each pattern of MSS with the exponent of 2. Furthermore, we have to consider the basic three parameters together with state transitions but not violate those rules. We define that PIi,j is the packet inter-arrival time of Ci,j. The frame duration F is assumed to be an exponent of 2ms. Firstly, we sort Ci,j of the same MSS i by delay bound of Ci,j
like 𝐷𝑖,1≤ 𝐷𝑖,2≤ ⋯ ≤ 𝐷𝑖,𝑎 and then choose the Di,1 and CAGE, which is the length of sleep cycle for Ci,1, qualified by (2).
2 𝑙𝑜𝑔2𝐷𝑖,1 = 2𝑛≤ 𝐷𝑖,1< 2𝑛+1
2 ≤ 𝐶𝐴𝐺𝐸 ≤2𝐹𝑛 (2) Secondly, we can set each length of sleep cycle for Ci,2…a like Ti,j.
𝑇𝑖,𝑗 = 𝑇𝑖,𝑗 −1× 𝐷𝑖,𝑗
𝑇𝑖,𝑗 −1× 𝐹 (3) Therefore, we can obtain each Ti,j, a multiple of CAGE. Assuming Pi,j is the expected packet size of Ci,j, we can compute the Then we schedule FRi,j continuously from the first frame. We can observe that each FRi,j is limited in the period of CAGE, just as a CAGE contains all FRi,j following (6).
𝐹𝑅𝑖.𝑗
𝑎𝑗 =1 ≤ 𝐶𝐴𝐺𝐸 × 𝐵𝑓 (6)
Figure 6. Example of PSS-CAGE
Finally, we can obtain a MSS pattern, which sleep cycle is CAGE, the listening period is 𝐹𝑅𝑖.𝑗 possible CAGE from (2) and calculate every possible sleep period.
We will choose the best CAGE of minimal power consumption.
The length of MSS pattern will be the exponent of 2, and we can use Algorithm 1 to check whether this MSS pattern can join to BS pattern.
4. SIMULATION RESULTS
We set up our simulation environment by referring to VoIP parameters and considering only downlink traffic. The length of an OFDM frame is assumed to be 4ms, and each MSS has four connections simultaneously. Our algorithm can still work by multiplying by five in (2) when the duration of frame is 5ms, even though it has slightly decreasing performance. Besides, our scheduling algorithm is able to gain much better performance in the simulation result when assuming that the duration of frame is an exponent of 2ms.
Each packet inter-arrival time of connections is picked between 20ms and 32ms randomly, and delay bound is chosen between 48ms and 200ms randomly. All of the simulation results are averaged from 100 rounds. Three metrics are used to evaluate the performance: total number of MSSs, sleep ratio and successful scheduling rate. The number of MSSs that can work simultaneously in the same BS is counted. The sleep period divided by the sleep cycle is sleep ratio. The successful scheduling rate is the ratio of the number of successful scheduling MSSs to the number of MSSs need to schedule.
4.1 Effects of Packet Size
In this simulation, we not only fix the Bf to 2500 bytes per frame but also randomly set packet size between 20 and 40 bytes per frame by increasing the interval gradually. In Fig. 7, we can observe that the number of MSSs decreases when the packet size increases, and PSS-CAGE can always maintain more MSSs simultaneously than PS. When packet size is much smaller between 20 and 40 bytes per frame, the number of MSSs by PSS-CAGE is almost 100% more than that by PS. This phenomenon happens because PSS-CAGE adopts the pattern calculated by an exponent of 2 with type II for each MSS. Unlike PS that always adopts the same minimal length of MSS pattern for each MSS, PSS-CAGE can use the maximum length of MSS pattern to schedule much more MSSs without overlapping or violating each delay bound. Furthermore, the results shown in Fig. 7 can response to the situation of Fig. 2.
Figure 7. Effects of packet size on the number of MSSs
4.2 Effects of Delay Bound
We fix the Bf to 2500 bytes per frame in this simulation and randomly pick the packet size of connection between 40 and 80 bytes per frame. We gradually loosen the delay bound of connection with a 25 frame increment. In Fig. 8, although PS can contain more MSSs than PSS-CAGE when the delay bound fixes in an integer but not randomly picked from an interval, the gap can be somewhat neglected. The reason is that maximum length of MSS pattern and minimum length of MSS pattern are the same in this situation, and PSS-CAGE can schedule each sleep cycle no more than that of PS. However, each sleep cycle of connection is not usually the same in the real world, so that we simulate some
situations randomly picked the delay bound from an interval. The simulation result presents that when the interval is more loosening, PSS-CAGE can contain more MSSs as a result of bigger gap between the maximum length of MSS pattern and the minimum length of MSS pattern. As the same, the results shown in Fig. 8 can response to the situation of Fig. 2.
Figure 8. Effects of delay bound on the number of MSSs
Figure 9. Effects of Bf on the sleep ratio
Figure 10. Effects of Bf on the successful scheduling rate
4.3 Effects of Maximum Bandwidth (B
f)
We randomly pick the packet size from 40 to 80 bytes per frame and gradually increased the Bf from 150 to 2500 bytes per frame.
Furthermore, we set five MSSs working simultaneously in this simulation. Through Fig. 9, we can observe that PS can not activate the sleep mode when Bf is lower than 500 bytes per frame because the total requested bandwidth of all MSSs exceeds Bf; nevertheless, PSS-CAGE can activate even though the Bf is very small. This phenomenon happens because PSS-CAGE can abandon some MSSs in order to successfully schedule others when Bf is very small; however, PS can only either successfully schedule all MSSs with all connections or schedule none of them.
In other words, PSS-CAGE gets better performance than PS when the BS lay in high loading. We can also observe that when Bf is lower than or equal to 500 bytes per frame, PSS-CAGE can successfully and completely schedule as many MSSs into BS as PS after Bf is greater than or equal to 750 bytes in Fig. 10.
Although PS still obtains higher sleep ratio when Bf exceeds 500 bytes per frame, PSS-CAGE is very close to PS as shown in Fig. 9.
The gap between PS and PSS-CAGE is only 3% and lower than pattern and does not violate its delay bound. In other words, PSS-CAGE can decrease additional listening period and try its best to approach the delay bound. According to Fig.7, the result indicates that PSS-CAGE can schedule more MSSs simultaneously than PS, and PSS-CAGE still maintains high sleep ratio as shown in Fig. 9.
6. ACKNOWLEDGMENTS
The authors would like to thank the National Science Council, Taiwan, Republic of China, under grant NSC96-2221-E-182-007-MY3.
7. REFERENCES
[1] IEEE Std 802.16-2009 “IEEE standard for local and metropolitan area networks part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems”, May 2009.
[2] C. So-In, R. Jain, and A.-K. Tamimi, “Scheduling in IEEE 802.16e mobile WiMAX networks: key issues and a survey”, IEEE Journal on Selected Areas in Communications, vol. 27, pp. 156-171, Feb. 2009.
[3] S.-L. Tsao and Y.-L. Chen,, “Energy-efficient packet scheduling algorithms for real-time communications in a mobile WiMAX system”, Computer Communications, vol.
31, pp. 2350-2359, June 2008.
[4] T.-C. Chen, Y.-Y. Chen, and J.-C. Chen, “An Efficient Energy Saving Mechanism for IEEE 802.16e Wireless MANs”, IEEE Transactions on Wireless Communications, vol. 7, pp. 3708-3712, October 2008.
[5] T.-C. Chen, J.-C. Chen and Y.-Y. Chen, “Maximizing unavailability interval for energy saving in IEEE 802.16e wireless MANs”, IEEE Transactions on Mobile Computing, vol. 8, pp. 475-487, April 2009.
[6] J.-J. Chen, S.-L. Wu and S.-W. Wang, “Power Saving Class Management for Energy Saving in IEEE 802.16e Wireless Networks”, in Proc. IEEE Int. Conf. Mobile Data Management, pp.490–495, Taipei, Taiwan, 2009.
[7] S.-C. Huang, R.-H. Jan, and C. Chen “Energy efficient scheduling with QoS guarantee for IEEE 802.16e broadband wireless access networks ”, in Proc. Int. Wireless
Communications and Mobile Computing Conf., pp. 547-552, Honolulu, Hawaii, 2007.
[8] S.-C. Huang, C. Chen, R.-H. Jan and C.-C. Hsieh, “An energy-efficient scheduling for multiple MSSs in IEEE 802.16e broadband wireless ”, IEEE Int. Symp. Personal, Indoor and Mobile Radio Communications, pp. 1-5, Cannes, France, 2008.
[9] W.-H. Liao and C. Liu, “An Efficient Scheduling Algorithm for Multiple MSSs”, National Computer Symposium, Taipei, Taiwan, 2009, in press.
[10] L.-H. Yen, C.-Y. Cheng, and C.-H. Wang, “Sleep scheduling that minimizes state transitions for IEEE 802.16e mobile subscriber stations”, Proc. Int. Conf. New Trends in
[10] L.-H. Yen, C.-Y. Cheng, and C.-H. Wang, “Sleep scheduling that minimizes state transitions for IEEE 802.16e mobile subscriber stations”, Proc. Int. Conf. New Trends in