Power-efficient and QoS-aware scheduling in ... - Semantic Scholar

Y.-I. Joo et al.: Power-Efficient and QoS-Aware Scheduling in Bluetooth Scatternet for Wireless PANs

1067

Power-Efficient and QoS-Aware Scheduling in Bluetooth Scatternet for Wireless PANs Yang-Ick Joo, Tae-Jin Lee, Doo Seop Eom, Yeonwoo Lee, and Kyun Hyon Tchah Abstract — In this paper, we propose a power-efficient and QoS (Quality of Service)-aware MAC (Medium Access Control) scheduling algorithm for Bluetooth scatternet. If the inter-piconet scheduler in a bridge node between piconets operates simply by the Round Robin (RR) policy, a number of time slots may be wasted due to the guard time for piconet switching and the exchange of POLL-NULL packets. To overcome the link wastage problem in Bluetooth scatternet scheduling, several algorithms have recently been proposed. Although they can provide all of the links of a bridge node with fair service opportunities, they may cause waste of wireless resources since different Bluetooth devices may have various traffic characteristics. In addition, since Bluetooth devices are often required to operate under limited battery capacity, the number of unnecessary piconet switching has to be minimized for the power-efficient operation of a Bluetooth scatternet. Therefore, we propose a mechanism to support the power-efficient operation of a Bluetooth scatternet while guaranteeing various QoS requirements of Bluetooth devices. The proposed algorithm is compliant with the current Bluetooth specification, and we demonstrate its improved performance via simulations.

maintains connections with multiple piconets in a time division manner. Bluetooth employs a Master-driven TDD (Time Division Duplex) scheme, and time slots are distributed alternatively between the Master and the Slave in a piconet. The Master can send packets to a Slave in even-numbered slots, while the Slave can only send packets to the Master in odd-numbered slots immediately after receiving packets from the Master [1]. In addition to the piconet scheduling, a scatternet has to maintain an inter-piconet scheduler to connect multi-piconets. If the inter-piconet scheduler in a PMP node operates simply according to the conventional Round Robin (RR) policy, a number of slots may be wasted due to frequent piconet switching and the exchange of POLL or NULL packets for signaling.

master slave slave/slave node master/slave node

Index Terms — wireless personal area networks, Bluetooth, scatternet, scheduling, and QoS I.

INTRODUCTION

(a)

The prosperity of wireless communication technology has been providing various new communication opportunities and services for personal use. A tremendous growth in popularity of wireless personal devices is increasingly requiring efficient communications between those heterogeneous devices. Although high-speed wireless LAN technology spreads out recently, Wireless Personal Area Network (WPAN) technology such as Bluetooth or ZigBee has its own merits and is continuously gaining interest for ubiquitous connection in home entertainment, security, and medical/military applications due to its inexpensive cost, low power consumption, and small size. The Bluetooth specification [1] defines two types of network configuration, i.e., the piconet and the scatternet. The piconet is the basic network unit, which is composed of a single master and up to seven active slave devices as shown in Fig. 1(a). However, it is possible for a single Bluetooth device to participate in more than one piconet as shown in Fig. 1(b). Such multi-piconet configuration is called the scatternet. A PMP (Participant in Multiple Piconets) node, i.e., bridge node, Contributed Paper Manuscript received August 6, 2003

(b)

Figure 1. Network configuration of Bluetooth (a) piconet and (b) scatternet.

To overcome the link wastage problem in Bluetooth scatternet scheduling, several algorithms [2]-[7] have been proposed. Among them, the scatter mode with credit scheme [2], [3] is promising and provides all of the links of a PMP node with fair service opportunities. However, since different Bluetooth devices may have various traffic characteristics, the uniform distribution of service opportunities is not an efficient policy from the perspective of obtaining the best possible QoS (Quality of Service) and the most efficient allocation of wireless resources. In addition, Bluetooth devices are often required to operate under limited battery capacity. Since guard time is used for piconet switching, it wastes transceiving power as well as processing power to re-synchronize whenever a piconet switching event occurs. Accordingly, we must minimize the number of unnecessary piconet switching events for the low-power operation of Bluetooth scatternets. Therefore, in this paper, we propose an efficient scatternet

0098 3063/00 $10.00 © 2003 IEEE

1068

IEEE Transactions on Consumer Electronics, Vol. 49, No. 4, NOVEMBER 2003

scheduling algorithm to support Bluetooth scatternet operations considering both power-efficiency and QoS requirements, which is conceived as one of the most important issues for the successful deployment of WPANs. This paper is organized as follows: In Section II, we describe the scatter mode with credit scheme [2], [3], the conventional inter-piconet scheduling scheme for Bluetooth systems. Our proposed algorithm to improve the conventional scheme is explained in Section III. In Section IV, computer simulations are executed and their results are discussed. Finally, in Section V, concluding remarks are presented.

links, i.e., one for each link’s credit account. And, the unused credits of a broken CE caused by POLL-NULL (P-N) sequences are equally redistributed to the other links. The original scatter mode with credit scheme presented so far may lead to frequent piconet switching events [3]. Since guard time is used for piconet switching, it is desirable to minimize the number of piconet switching events in order to save power consumption. In order to compensate for the problem, the abortion of an ongoing CE is allowed only if the credits of another link exceed the credits of the current communicating link by a certain threshold, Nswitch_th [3].

II. SCATTER MODE FOR BLUETOOTH SCATTERNET SCHEDULING

III. THE PROPOSED SCHEDULING POLICY USING THE OPTIMIZED Nswitch_th

The scatter mode [2], [3] is a new connection mode for Bluetooth and has been proposed to support Bluetooth scatternet operation based on time-division sharing of resources. Although the scatter mode is very similar to the sniff mode, the scatter mode is not intended for the purpose of power saving [3]. The mechanism defines two terms, Presence Point (PP) and Communication Event (CE). A PP is a rendezvous point at which the master of a piconet and a PMP node meet, and it is composed of two consecutive slots. When the master wants to communicate with the slave (PMP node) in the scatter mode, the master sends a signaling message to the slave in the first slot of a PP. A slave that wants to communicate with the master will listen to its master in this slot. If the slave receives the signaling message, it will reply with a corresponding signaling message in the subsequent slot [2], [3]. Once two devices have successfully exchanged signaling messages at a PP, a CE starts at the master-to-slave slot subsequent to the PP. During a CE, both devices stay in the same piconet and may communicate with each other. Even though the scatter mode is not initially intended for power saving, efficient inter-piconet scheduling for the scatter mode will be very helpful. So the scatter mode employs the credit scheme as its [HH1] inter-piconet scheduling algorithm [2], [3]. In order to assign a priority to each link associated with each piconet in a PMP node of a scatternet, a counter is maintained for each of these links, whose value is debited or credited depending on the link’s utilization. In addition, a temporary account is maintained for a PMP node. If the counter of another link at the upcoming PP has a higher credit, the ongoing CE is aborted. Once a slot of a link is used for communication, the credit account of the link is automatically decreased by one. In order to keep the sum of all the credits among all the links constant, one credit per slot should be increased in the temporary account for the PMP node while one credit of the polled link is decreased in the corresponding account. As soon as the temporary account reaches M credits, where M is the number of links of the PMP device, they are distributed equally amongst all of the credit accounts of the M

For efficient MAC scheduling, QoS requirements of each applications or upper layers must be taken into consideration in the MAC scheduler of a scatternet. Therefore, QoS-based MAC scheduling for Bluetooth systems using QoS manager as shown in Fig. 2 can greatly improve system performance and efficiency. The QoS manager needs to find appropriate scheduling parameters and to optimize them of the MAC scheduler by reflecting QoS requirements from its upper layers or events from external environment. Application 1

Application 2

Application n

QoS negotiations

Parameter Optimization Optimized parameters

Bluetooth Piconet / Scatternet MAC Scheduler

Events from external environment

QoS Manager

Figure 2. QoS-aware MAC scheduling framework. Presence Point Transmit or Listen

nTpp,i Link i

ci

1  × nT pp ,i  M 

ci +  Link j

cj

Link k piconet switching

ck

c j − nTpp ,i +

 1  M × nT pp ,i 

t

 1  ck +  × nTpp ,i  M 

Figure 3. Variation of credits at each piconet link in a scatternet to compute Nswitch_th,i and to support QoS requirements.

In [3], it is recommended that Nswitch_th should be set to Tpp, an interval between two consecutive PPs in slots, which is fixed at 16. The value Nswitch_th and the other parameters are required to have fixed values. Thus, the scatter mode with basic credit scheme [2], [3] only focuses on the fairness among


the connected links of a PMP node. In order to improve the power-efficiency and to satisfy the QoS requirements of a Bluetooth scatternet, we propose to change the Nswitch_th adaptively according to the polling interval of the scatternet, Tscatter_poll, which is the QoS parameter in the MAC layer of a Bluetooth scatternet. In Fig. 3, the scatter mode with credit scheme is illustrated to explain the behavior. Assuming the credit scheme is adopted for scatternet scheduling, for link i to be serviced at the nth PP after the piconet switching event from link k to link j, following equation (1) regarding the credits must hold.

1069

may be some timing difference between Tscatter_poll,i and the nTpp,i. This gap can be overcome by substituting Tscatter_poll,i - ti with Tscatter_poll,i - ti - Tpp,i in (3) since the difference will then be more conservative by the amount of Tpp,i. Thus, we will use this term in the sequel. Hence the requirement Tscatter_poll,i can be met automatically, and it becomes unnecessary to check the scatternet polling interval. Scatter mode Basic credit scheme in [2], [3]

1  1  c i +  × nT pp ,i  > c j − nT pp ,i +  × nT pp ,i  + N switch _ th , i , (1) M  M 

where M is the number of links of the PMP node, ci and cj are the credit values of link i and j, respectively, when the piconet switching event from link k to link j has occurred, T pp ,i is the time interval between two successive PPs, j

= arg max cm and

if (ci - cj > Nswitch_th,i) or POLL-NULL seq. exchange piconet_switching to link i update Nswitch_th,i by (3) end

Figure 4. Summary of our proposed algorithm.

1≤ m ≤ M

x  denotes the largest integer which is less than or equal to

IV. SIMULATION RESULTS

x. In order to guarantee the QoS requirement of link i, the parameter, Tscatter_poll,i, has to be taken into consideration as follows.

(

)

1  c i +  × Tscatter _ poll ,i − t i  > M  1  c j − T scatter _ poll ,i − t i +  × Tscatter _ poll ,i − t i  + N switch _ th ,i M 

(

)

(

)

Piconet2

Piconet1

flow 1

S1 M

S1 M

S3/S3/S3

(2) flow 2 S2

S2

flow 4 flow 3

where ti is the elapsed time since the most recent CE of link i. Since Nswitch_th,i is an integer value, the following relation can be derived from (2) assuming both sides are equal:

(

)

N switch _ th ,i = Tscatter _ poll ,i − t i − c j − c i  .

(3)

Note that Nswitch_th,i is proportional to Tscatter_poll,i in (3). This relation is reasonable since a link with a larger Tscatter_poll,i value requires fewer rendezvous, and vice versa. Therefore, the proper value of Nswitch_th can prevent unnecessary frequent piconet switching and take the QoS requirement of each link into consideration. In Fig. 4, we summarize the process of the proposed scheduling mechanism. The proposed scheme maintains the framework of basic credit scheme and updates the threshold value Nswitch_th,i as determined by (3) when a piconet switching event occurs. By applying the mechanism, a scatternet can operate in a more resource-efficient manner since it can reflect the QoS requirements of the piconet links. Note that the mechanism is very simple, but nevertheless provides an intimate mapping between the QoS requirements at the application layer and the parameters at the MAC layer. There

M S1 S2

Piconet3

Figure 5. WPAN topology for the simulation of the proposed scatternet scheduling mechanism.

The performance of the proposed algorithm is evaluated via simulations. Fig. 5 shows the scatternet topology used in the simulation. Three piconets are connected each other by a PMP node, which is a slave node for all of the linked piconets. Tpp is fixed at 16 as in [3]. Since it is recommended that Nswitch_th should be set to Tpp in [3], we consider various values of Nswitch_th including 16 and the value determined by (3). For three piconet links of the PMP node, different Tscatter_poll,i values are assumed to support different QoS requirements, i.e., Tscatter_poll,i = 60, 120, and 180 slots for link i=1, 2, and 3, respectively. Four intra- and inter-piconet traffic flows are illustrated in Fig. 4. It is assumed that only single slot packets are allowed for transmission, and the packet generation process on each traffic flow is assumed to be a Poisson process.


number of passed packets per 2000 slots

In order to minimize the effects of piconet switching due to POLL-NULL sequences, we set the average packet arrival rate to one packet per slot.

560 540 520 500 480 460 440 420 400 380 360 340 320 300 280 260 240 220 200

Nswitch_th,opt Nswitch_th=10 Nswitch_th=16 Nswitch_th=20 Nswitch_th=30

1

2

3

4

5

6

7

8

9

time [s]

Figure 6. Throughput variations of link 1.

In Fig. 9, such differentiated allocation of time slots is well observed for flow 1, which traverses link 1 and link 2. By adopting the proposed scheme, flow 1 passing through link 1 obtains more bandwidth, since link 1 with more inter-piconet traffic and the tightest QoS requirement, i.e., the shortest Tscatter_poll, is assigned more service slots than the others to satisfy its higher QoS level.


1070

560 540 520 500 480 460 440 420 400 380 360 340 320 300 280 260 240 220 200


1

2

3

4

5

6

7

8

9

X Axis Title


500 450 400

260

350


300 250 200 1

2

3

4

5

6

7

8

9

time [s]


Fig. 6 ~ Fig. 8 represent the measured number of packets traversing each link, where each link is associated with the corresponding piconet. When the conventional credit scheme with a fixed Nswitch_th, are employed, the throughput of link 2 (piconet 2) is shown to be relatively higher due to its traffic on both directions, i.e., flow 1 and flow 2, as shown in Fig. 7. On the contrary, link 1 (piconet 1) and link 3 (piconet 3) show lower throughput since competing flows, i.e., flow 1 and flow 4 for link 2, and flow 3 and flow 4 for link 3, are quenched at the links as shown in Fig. 6 and Fig. 8. However, when the QoS requirement of each link is taken into consideration by using the optimum Nswitch_th determined by (3), we can achieve improved performance for link 1 and link 2, although the throughput performance for link 3 is slightly degraded due to its rather loose QoS requirement.



550

240


220

200

180

160

1

2

3

4

5

6

7

8

9

time [s]

Figure 9. Throughput variations of flow 1.

In order to evaluate the power-efficiency of the proposed scheme, we plot the number of piconet switching events, as shown in Fig. 10. As expected, the lower values of Nswitch_th causes more frequent piconet switching, and vice versa. On the other hand, use of the optimal Nswitch_th determined by (3), reduces the number of piconet switching events. Therefore, the proposed method can improve the power-efficiency of the Bluetooth scatternet. Although the case where Nswitch_th=30 results in the smallest number of piconet switching events, it is not able to satisfy the QoS requirements as described earlier.

number of piconet switchings per 2000 slots


power consumption by reducing the number of unnecessary piconet switching events. Such a power-efficient and QoSaware scheduling mechanism will play a key role in a typical personal/mobile wireless communication environment, e.g., WPANs for applications where battery power is limited and radio resource is scarce. Since the additional overhead for the computation is very small, the proposed method can easily be applied to the current Bluetooth specification [1] and the scatter mode [2], [3].

85 80

Nswitch_th,opt Nswitch_th =10 Nswitch_th =16 Nswitch_th =20 Nswitch_th =30

75 70 65 60

1071

55 50

REFERENCES

45 40 35 1

2

3

4

5

6

7

8

9

time [s]

Figure 10. Number of piconet switching events.

Finally, the actual ratios of the allotted service time slots among the links are measured in order to check if the QoS requirements of the links are met. If the service opportunities are optimally distributed according to Tscatter_poll, time slots are to be shared following the ratio of the reciprocals of Tscatter_polls, i.e., the ratios of time slots for link 1, link 2, and link 3 are 6, 3, and 2, respectively. While the results of the basic credit scheme with a fixed Nswitch_th show uniform distribution of service opportunities to the links, those of our scheme with Nswitch_th,opt, i.e., optimal value of Nswitch_th, indicate the proportional distribution with inverse proportion of the ratio of Tscatter_poll values. Thus, each link’s QoS requirement can be satisfied by our proposed mechanism.

link 1 link 2 link 3

Nswitch_th=30 = 30 Nswitch_th

= 20 Nswitch_th Nswitch_th=20

[1] Bluetooth Special Interest Group, Specification of the Bluetooth system Version 1.1B, Specification Vol. 1&2, Feb. 2001. [2] Bluetooth Special Interest Group, Scatter mode, PAN IPS Whitepaper, Ver. 0.25, August 2001. [3] Bluetooth SIG Radio WG [BT PAN IPS], Improvement proposal for scatternets, Ver. 0. 46, April 2002. [4] A. Rácz, G. Miklós, F. Kubinszky, and A. Valkó, “A pseudo random coordinated scheduling algorithm for Bluetooth scatternets,” in Proc. of MobiHOC, pp.193-203, 2001. [5] N. Johansson, F. Alriksson, and U. Jönsson, “JUMP mode-a dynamic window-based scheduling framework for Bluetooth scatternets,” in Proc. of MobiHOC, pp.204-211, 2001. [6] S. Baatz, M. Frank, C. Kühl, P. Martini, and C. Scholz, “Adaptive Scatternet Support for Bluetooth using Sniff Mode,” in Proc. of LCN, pp.112-120, 2001. [7] P. Johansson, M. Kazantzidis, R. Kapoor, and M. Gerla, “Bluetooth: An Enabler for Personal Area Networking,” IEEE Network, Sept./Oct. 2001, pp. 28-37.

Yang-Ick Joo received the B.S. and M.S. degrees in Electronics Engineering from Korea University, Seoul, Korea in 1998 and 2000, respectively. He has been in the Ph.D. program in Electronics Engineering at the same university, since 2000. His research interests include PHY and MAC layer solutions for wireless system, Bluetooth, wireless PAN, and ubiquitous networking.

Nswitch_th=16 = 16 Nswitch_th

Nswitch_th=10 Nswitch_th = 10

Nswitch_th,opt Nswitch_th,opt optimal(6:3:2) Optimal (6:3:2) 0

10

20

30

40

50

60

70

80

90

100

service allocation rate [%]

Figure 11. Proportion of time slots allotted to the links.

V. CONCLUSION In this paper, we have proposed the optimization process of MAC scheduling parameter for an efficient Bluetooth scatternet operation. The proposed scheme can satisfy each piconet link’s QoS requirement, and at the same time, lowers

Tae-Jin Lee received his B.S. and M.S. in Electronics Engineering from Yonsei University, Seoul, Korea in 1989 and 1991, respectively, and his Ph.D. in Electrical and Computer Engineering from The University of Texas at Austin, Austin, TX in 1999. In 1999, he joined Corporate R&D Center, Samsung Electronics where he was a senior engineer. Since 2001, he has been an Assistant Professor in the School of Information and Communication Engineering at SungKyunKwan University, Korea. His research interests include design and performance evaluation of communication networks and systems, wireless PAN, Bluetooth, wireless LAN, and ad-hoc networks. Doo Seop Eom received the B.S. and M.S. degrees in Electronics Engineering from Korea University, Seoul, Korea in 1987 and 1989, respectively. In 1999, he received the Ph.D. degree in Information and Computer Sciences from Osaka University, Osaka, Japan. He joined the Communication Systems Division, Electronics and Telecommunications Research Institute (ETRI),

1072


Korea, in 1989. From September 1999 to August 2000, he was an Associate Professor of Wonkwang University, Korea. Since September 2000, he has been an Associate professor in the Department of Electronics Engineering at Korea University. His research interests include communication network design, Bluetooth, ubiquitous networking, and Internet QoS. Yeonwoo Lee is currently a research fellow with the School of Electronics and Engineering at the University of Edinburgh since 2000. He received a MS and Ph.D. from Department of Electronics Engineering with Korea University in 2000. His research interests are in wireless multimedia mobile telecommunication systems, radio resource management, radio resource metric estimations, and particularly their applicable issues to 3G and beyond mobile communication systems.

Kyun Hyon Tchah received the B.S. and Ph.D. degrees in Electrical Engineering from Seoul National University, Seoul, Korea in 1965 and 1976, and M.S. degree in Electrical Engineering from Illinois University. Since 1978, he has been a Professor in the Department of Electronics Engineering at Korea University. He was the President of the Korean Institute of Communication Sciences (KICS) from January 1998 to December 1998, and the chairman of IEEE Seoul Section from 2001 to 2002. He has been the Fellow in IEE since 2000. His current research interests are the areas of PHY and MAC layer solutions for wireless communication systems, wireless LAN/PAN, and ubiquitous networking.