Chapter 2 Reviews and Related Works
2.2 Power saving protocols for 802.16e WMANs
IEEE 802.16e standard [8] defines three power saving classes shown as figure 2-3.
Each connection of a MSS can select a specific power-saving class. MSSs need to negotiate with the BS to decide the power-saving parameters such as listening period and sleeping period. A round of a sleeping period and a listening period is defined as one sleeping cycle.
Figure 2-3 the three power saving classes of IEEE 802.16e WMANs
In the first power-saving class, each MSS sleeps for a period of time and then wakes up to listen. During the listening period, if no packets are sent or received, the MSS doubles the length of its next sleep cycle. This kind of power-saving is suitable for web browsing or data access services. In the second power saving class, a MSS is required to repeat the sleeping and listening periods in a round-robin fashion. The length of both the sleeping and listening periods in the sleeping cycle is fixed. This sleep mode works well for real-time applications that have packets to send or receive periodically such as VoIP and video streaming services. The third power-saving class requires a predefined sleep period length. The MSS simply sleeps for a predefined period of time and then returns to normal operation.
Several studies have been proposed to analyze the power consumption for IEEE 802.16e while a MSS operates in the power-saving mode [48-52]. There are some scheduling mechanisms proposed to determine the length of sleeping period [36-38]. In [36], the length of the sleeping period is varied according to the traffic type. However, the
scenario is only valid under one MSS and the QoS delay constraint is not considered. In [37], although the QoS delay constraint is considered, the scenario can not be applied to multiple-MSS environment. In [38], a scheduling algorithm for multiple MSSs with QoS delay constraints is proposed. Authors classified the MSSs into two categories: primary and secondary MSSs. In order to save energy, the algorithm grants one primary MSS to use the bandwidth in burst mode. The other MSSs or we say secondary MSSs are only given the necessary bandwidth to meet the requirements of the QoS delay constraint.
However, this algorithm works well only when all MSSs have light traffic load. In [39], an analysis is provided via semi-Markov Decision Processes (Semi-MDP) to find an optimal way to switch between Type I and II power saving modes while considering the mix modes operation. However, the discussion of bandwidth utilization is absent. It only considers how to select a suitable power saving class.
Thus, in this dissertation, we consider the UGS traffic which has QoS delay constraint and adapt the second power-saving class defined in IEEE 802.16e standard.
We propose two kinds of power saving schedule approaches for multiple MSSs, the periodical autonomic sleeping cycle approach (PASC) and the periodical uniform sleeping cycle approach (PUSC). We will discuss them in the chapter 5.
Chapter 3
Power Saving Protocols for IEEE 802.11 Infrastructure Mode
This chapter presents a method to arrange the sleep schedule of stations in the infrastructure mode of IEEE 802.11 wireless local area networks (WLANs). The goal is to balance the number of wakeup stations in each beacon interval. This method reduces the probability of collision and thus the station can save more power. To avoid the contention, this method considers how to poll the wakeup stations to send the PS-Poll frame to get back their buffered data. Three different access scheduling mechanisms are proposed for the contention avoidance. In the first mechanism, only one wakeup station is scheduled to access the buffered data. The second and third mechanisms based on the smallest association ID (AID) first and the smallest queue length first, respectively, arrange a subset of wakeup stations to get back their buffered data within a beacon interval. Simulation results show that the proposed methods are effective in the power-saving.
The organization of this chapter is as follows. Problem statement is given in section 3.1. A new wakeup scheduling mechanism is considered in section 3.2 and the three contention avoidance mechanisms for polling wakeup stations are presented in section
3.3. In section 3.4, we give the simulation results to show the effectiveness of our proposed methods. Finally, a summary is given in section 3.5.
3.1 Problem statement
Considering the figure 3-1, there are six power-saving stations, A, B, C, D, E, and F, whose listen-intervals are 1, 2, 3, 6, 6, and 6. wi(t) is used to indicate the sleeping state of station i at beacon interval t. If wi(t) is 1, station is active, otherwise station is in sleep mode. n(t) is the total number of wakeup stations in beacon t.
Figure 3-1 a sequence of wi(t) and n(t), n(t)J : the n(t) after J is included In order to trace the wakeup time of each station, AP needs to maintain a wakeup counter, denoted as , for each sleeping station i. indicates the remaining beacon interval that station i will wake up. Initially, AP sets =
) listen-interval of station i. Station i wakes up if become 0 and the counter will
reset to = for further counting down. AP sets counter for beacon interval t+1 as follows:
⎩⎨
The wakeup scheduling problem (WSP) can be formulated as follows: Given a set of sleeping stations S at beacon interval t. Consisting m stations and each of the station i has wakeup count and listen-interval . For a new sleeping station j, we assign an initial value to such that the maximum value of
= 1, 2, 3 . . . is minimized. In the next section, we present the mechanism of load-aware wakeup scheduling.
3.2 Load-aware wakeup scheduling
By observing the sequence of n(t) in Figure 3-1, we find that a pattern repeats every six beacon intervals, e.g., (n(1), n(2), …, n(6)) = (n(7), n(8), …, n(12)) = (n(13), n(14), …, n(18)) = (3, 2, 1, 3, 2, 3). The length of this repeating pattern, r, can be found by computing the least common multiple (lcm) of listen-interval , i∈S. For example, the listen-intervals of stations, A, B, C, D, E, and F are 1, 2, 3, 6, 6, and 6, respectively, in
li
Figure 3-1. Then
r = lcm {1, 2, 3, 6, 6, 6} = 6
Thus, n* = max{n(t + 1), n(t + 2), …, n(t + r)} = max{n(t + k) | k = 1, 2,…}. Now,
we want to add a new sleeping station j with listen-interval to the sleeping station set S with repeating pattern size r and assign an initial value to c
lj
j(t). A stepwise solving method for the WSP problem is given as follows.
1. Find r = lcm{ , r} and a sequence of total number of wakeup stations (n(t + 1), n(t + 2), …, n(t + r)) for the first r intervals for sleeping station set S ∪{j}.
li
2. For i = lj − 1, …, 1, 0, perform the following operations:
(a) Set cj(t) = i and find (wj(t + 1), wj(t + 2), …, wj(t + r));
(b) Set (n(t + 1), n(t + 2), …, n(t + r)) = (n(t + 1), n(t + 2), …, n(t + r)) + (wj(t + 1), wj(t + 2), …, wj(t + r));
(c) Find ni = max{n(t + 1), n(t + 2), …, n(t + r)}.
3. Find n* = min{ni | i = lj − 1, lj − 2, …, 0}, say n* = nk, and thus set cj(t) = k.
For example, six stations with r = 6 as given in Figure 3-1, station J with = 3 enters the sleeping mode. The AP applies the above solving method to determine the initial value of counter c
lJ
J(t) for station J as follows.
1. r = lcm{3,6} = 6 and (n(t + 1), n(t + 2), …, n(t + 6)) = (3,2,1,3,2,3) 2. i = 2:
(a) Set cj(t) = 2 and find (wj(t + 1), wj(t + 2), …, wj(t + 6)) = (0,0,1,0,0,1);
(b) Set (n(t + 1),n(t + 2), …, n(t + 6)) = (3,2,1,3,2,3) + (0,0,1,0,0,1) = (3,2,2,3,2,4);
(c) Find n2 = max{3,2,2,3,2,4} = 4.
i = 1:
(a) Set cj(t) = 1 and find (wj(t + 1), wj(t + 2), …, wj(t + 6)) = (0,1,0,0,1,0);
(b) Set (n(t + 1), n(t + 2), …, n(t + 6)) = (3,2,1,3,2,3) + (0,1,0,0,1,0) = (3,3,1,3,3,3);
(c) Find n1 = max{3,3,1,3,3,3} = 3.
i = 0:
(a) Set cj(t) = 0 and find (wj(t + 1), wj(t + 2), …, wj(t + 6)) = (1,0,0,1,0,0);
(b) Set (n(t + 1), n(t + 2), …, n(t + 6)) = (3,2,1,3,2,3) + (1,0,0,1,0,0) = (4,2,1,4,2,3);
(c) Find n0 = max{4,2,1,4,2,3} = 4.
3. Find n* = min{4,3,4} = 3, i.e., n* = n1, and thus set cj(t) = 1.
Note that according to IEEE 802.11 standard, if mobile station j has no data to send, it can send a Null data frame with Power Management bit set to 1. The AP begins to buffer frames and sends an ACK frame to the station after receiving the Null data frame.
We can just modify this step to incorporate our wakeup scheduling in IEEE 802.11 standard as follows: The AP begins to buffer frames, determines cj(t) and sends an ACK frame with cj(t) value to station j after receiving the Null data frame. Then, the station j sets its wakeup counter to cj(t) and enters the sleeping mode.
3.3 Contention avoidance traffic scheduling
In the previous section, we arrange stations’ wakeup beacon intervals so that the number of wakeup stations in each beacon interval is balanced. In this section, we consider how to inform stations that frames are buffered such that the contention is
avoided. Three different access scheduling mechanisms are proposed for the contention avoidance problem. In the first mechanism, only one wakeup station is scheduled to access the buffered data in a beacon interval by marking one bit in TIM. The second and third mechanisms schedule multiple wakeup stations to get back their buffered data within a beacon interval. The access sequence within the beacon interval is according to their AIDs and the length of queuing data.
3.3.1 Multiple wakeups single access
One of the simple ways to avoid contention is that we only inform a station that AP has its buffered frames at each beacon interval. So there is no contention problem of sending PS-Poll frame to get back its buffered data. Let Sw(t) be the set including all stations waking up at beacon interval t. That is,
Sw(t)={i|i∈S, ci(t)=0}
where S is the set including all sleeping stations. Let Sb(t) be the set including all stations that frames are buffered in AP at beacon interval t. Thus, we can choose a station, say station v, from set Sw(t) ∩ Sb(t) with a largest listen-interval to inform that the AP has buffered frames for it.
lv
It is possible that some stations in set Sw(t) ∩ Sb(t) use small listen-interval and they are never chosen by AP. To avoid such a case, we associate each station v in Sw(t) ∩ Sb(t) with an age, denoted as av. Initially, the age of each station is set to zero. For each beacon interval, if a station in set Sw(t) ∩ Sb(t) is not selected to inform, AP increases its age by
one; otherwise, AP sets its age to zero. Thus, AP can choose a station, say station v, from set Sw(t) ∩ Sb(t) with a largest value of lv + av to inform. Here we denote lv + av as pv.
Figure 3-2 shows an example of this mechanism. Consider that there are four stations, A, B, C, and D, with listen-interval ( , , , ) = (2, 2, 3, 1). Suppose there are 1,1,1,1 packets send to station A, B, C, and D in every beacon interval. Packet arrival rate of each station is one frame per beacon interval. In beacon interval t, stations A, C, and D wake up in which station C, has maximum p
lA lB lC lD
C = lC + aC = 3 + 0, is indicated in TIM to inform it that the AP has buffered its data. Stations A and D are deferred to their next wakeup beacon intervals. The AP sets ages aA = aA + 1 and aD = aD + 1. In the beacon interval t + 1, stations B and D wake up. Because
B + a
l B = B lC + aC = 2, the AP selects station B, arbitrarily, to inform it has buffered frames. Similarly, station A is chosen to inform in beacon interval t + 2. At beacon interval t + 3, lB + aBB < lC + aC < lD + aD and thus station D is chosen to inform.
3.3.2 Multiple wakeups multiple accesses
Although the multiple wakeups single access mechanism avoids the contention among stations, it may lower the bandwidth utilization and increase the transmission delay. However, the AP knows how many frames it has buffered in queue, transmission rate and the length of beacon interval. Thus, the AP can determine how many frames it can transmit in a beacon interval and schedule the buffered frames by means of announcing the TIM. In the following, we give two methods to arrange the access sequence of stations.
3.3.2.1 The smallest AID first
Figure 3-3 an example of smallest AID first method
In order to control the traffic load in a beacon interval, AP selects a set of stations with an appropriate size from Sw(t) ∩ Sb(t) to inform them to retrieve the data. That is the total amount of buffered frames of selected stations should be less than the capacity of a
not get back its buffered data. Next, we modify the power management scheme of IEEE 802.11 WLANs such that a station retrieves the buffered frame according to the sequence of AID marked in TIM. That is, the station with smallest AID among the selected stations sends PS-Poll frame to retrieve buffered data first.
Figure 3-3 shows an example of the AID sequence method. There are four stations, A, B, C, and D with listen-interval ( , , , ) = (2, 1, 3, 2) with under the service of an AP. Suppose there are 2, 2, 1, 2 packets send to A, B, C, D in each beacon interval. Their corresponding AIDs are 1, 2, 3, and 4 for stations A, B, C, and D, respectively. Suppose packet arrival rates of stations A, B, C, and D are 2, 2, 1, and 2 per beacon interval. The maximum number of frames that AP can transfer to stations in a beacon interval is 8. In beacon interval t, all of these four stations wake up. Because the number of buffered frames is 2 + 2 + 1 + 2 = 7 (7 < 8), the AP marks AIDs 1, 2, 3, and 4 in the TIM. The stations check the TIM in beacon frame. They learn that 4 stations will send PS-Poll to retrieve their buffered frames and every station knows which station precedes it in access sequence. For example, station C has to wait stations A and B finishing their access. In beacon interval t + 2, S
lA lB lC lD
w(t + 2) ∩ Sb(t + 2) = {A, B, D} and the number of frames buffered for stations, A, B, and D is 10 (10 > 8). Thus, based on the values of pA and pD, the AP selects stations A and D to inform them to retrieve the buffered data.
3.3.2.2 The smallest queue length first
Instead of the smallest AID first, the AP can arrange the access sequence for the
queue length receives a highest precedence and thus it can have a longer sleeping time. In this method, we need to add an information element, describes the access sequence, as a component of the beacon frame. The station checks this information element for the access sequence.
Figure 3-4 an example of the smallest queue length first method
Figure 3-4 shows an example of the smallest queue length first method. There are three stations, A, B, and C with listen-interval ( , , ) = (2, 1, 2) with under the service of an AP. Suppose packet arrival rates of stations A, B, and C are 2, 2, and 1 frames per beacon interval. The maximum frame size that AP can transfer to stations in a beacon interval is assumed to be 8. In beacon interval t, all of the three stations wake up.
Because the number of buffered frames are 2+2+1=5 (5<8), the AP marks AIDs 1, 2, and 3 in the TIM and adds the access sequence C, A, and B in the beacon frame. In beacon interval t+2, the access sequence is C, B, and A. Note that if two stations have same queue length, AP uses their p
lA lB lC
v values to break the tie.
3.4 Simulation and results
3.4.1 Performance metrics and environment setup
In this section, we show the performance analysis for the proposed schemes:
1. Load-aware wakeup scheduling (LAWS);
2. LAWS with multiple wakeups single access (LAWS+MWSA);
3. LAWS with multiple wakeups multiple access and the smallest AID first (LAWS+SAF);
4. LAWS with multiple wakeups multiple access and the smallest queue length first (LAWS+SQLF).
Note that all of these four schemes are enhanced from the PS mode of 802.11. The LAWS arranges station’s wakeup time. The MWSA, SAF, and SQLF schemes can be used by AP to schedule the access sequence by marking the bits in TIM. We compare their performances against pure IEEE 802.11 PS mode by simulation. The performance metrics are given as follows:
1. Average sleeping time of the station: This measure is the duration that a station stays in the sleeping mode. If a scheme can make stations stay more time in sleeping, then stations will save more power.
2. Average throughput: This value shows the total amount of data successfully transmitting per second. If AP can efficiently schedule and distribute the access of its serving stations, it will have higher data throughput.
3. Average latency of a successful transmission: The latency is defined as the time duration starting while a packet is issued and buffered at AP and ending when the target station returns the acknowledge. An AP with a good scheduling scheme will make the latency as small as possible. Thus, the resources required for buffering data can be reduced.
Table 3-1 detail simulation configurations
Data rate 11Mbps
MAC header 28 bytes IP header 20 bytes UDP header 20 bytes Beacon frame 28 bytes ACK frame 14 bytes PS-POLL frame 14 bytes
SIFS 0.00001 sec
DIFS 0.00005 sec
Slot time 0.00002 sec Beacon interval 0.1 sec
Our simulation uses an IEEE 802.11b wireless communication module with 11 Mbps data rate. An AP can serve at most 30 stations. Each station will randomly set 1 to 5 beacon intervals as its listen-interval size and its packet arrival rate is 3 packets per beacon interval. Packet size in our simulation is fixed and set to 1 Kbytes.
Communication channel assumes to be clear and symmetric. The total simulation time is 3 minutes. The details of other simulation configurations such as header length, and inter-frame spaces (IFS) are listed in Table 3-1. Simulation results will compare the IEEE 802.11 PS mode with the proposed LAWS, LAWS+MWSA, LAWS+SAF, and LAWS+SQLF schemes.
3.4.2 Results and discussion
Sleep time (sec)
Figure 3-5 the average sleeping time
Figure 3-5 shows the relation between average sleeping time and number of stations.
Considering contention-based schemes, LAWS can have more sleeping time than IEEE 802.11 PS mode in any size of stations. By using LAWS+MWSA, LAWS+SAF, and LAWS+SQLF schemes to reduce the contention within a beacon interval, stations can have more time on staying in sleeping than LAWS and IEEE 802.11 PS mode. In this figure, it seems that LAWS+MWSA has better sleeping time than LAWS+SAF and LAWS+SQLF. However, we will find in figure 3-7 that it trades the transmission latency with the sleeping time.
Figure 3-6 shows the average throughput for each scheme. From this figure, we can explicitly find that the throughput of IEEE 802.11 PS mode falls down when station number is greater than 20. However, our proposed schemes, LAWS, LAWS+MWSA, LAWS+SAF, and LAWS+SQLF, are not influenced as number of station increases. This is because our schemes can efficiently avoid the data collision between stations.
Throughput (bps)
Figure 3-6 the average throughput for each scheme
In Figure 3-7, we show the average latency of a successful transmission for each scheme. For LAWS, LAWS+SAF, and LAWS+SQLF, all of their latency is smaller than 0.3 sec and increase slowly as number of stations grows. Because only one station is indicated within a beacon interval, the latency of LAWS+MWSA scheme is longer than the other proposed schemes. The pure IEEE 802.11 PS mode, however, will suffer the worst latency while number of stations increases.
Finally, Figure 3-8 shows the improving rate of sleeping time for each proposed scheme (compared to pure IEEE 802.11 PS mode). The improving rate Ri of scheme i is
defined as 100%
0 0 ×
= − S
S
Ri Si , where S0 and Si are the average sleeping times for pure
IEEE 802.11 PS mode and the proposed scheme i, respectively. By efficiently scheduling the wakeup time of sleeping stations, the sleeping duration of LAWS, LAWS+MWSA, LAWS+SAF, and LAWS+SQLF schemes can be improved significantly.
Figure 3-7 the latency of a successful transmission for each scheme
Improving rate (%)
Figure 3-8 the improving rate of sleeping ratio
3.5 Summary
In this chapter, we propose a load-aware wakeup schedule scheme for infrastructure mode of IEEE 802.11 WLANs. The LAWS scheme balances the number of wakeup stations in each beacon interval to reduce the amount of contention stations. For avoiding
of the wakeup stations within a beacon interval. Simulation results show that comparing to IEEE 802.11 PS-mode, the proposed LAWS, MWSA, SAF, and SQLF schemes can efficiently improve the sleeping duration of each station, average throughput, and transmission delay.
The following two issues should be considered in the implementation of the proposed schemes:
1. An aging function should be implemented in the AP to determine when buffered frames are old enough to be discarded.
2. If the mobile station misses the beacon, it should remain awake until it receives the next beacon. The mobile station checks the beacon frame. If the bit corresponding to its AID is set to zero in the TIM, or else it has retrieved all buffered frames, the mobile station can resume the sleeping mode by asking AP for a new wakeup counter cj(t). In the LAWS+SAF and LAWS+SQLF schemes, the mobile station misses the beacon can not show up to retrieve the buffered data in its turn. The next station in the access sequence can send PS-Poll frames to get back its buffered data if it finds that the medium has been
2. If the mobile station misses the beacon, it should remain awake until it receives the next beacon. The mobile station checks the beacon frame. If the bit corresponding to its AID is set to zero in the TIM, or else it has retrieved all buffered frames, the mobile station can resume the sleeping mode by asking AP for a new wakeup counter cj(t). In the LAWS+SAF and LAWS+SQLF schemes, the mobile station misses the beacon can not show up to retrieve the buffered data in its turn. The next station in the access sequence can send PS-Poll frames to get back its buffered data if it finds that the medium has been