㪧㫉㫆㪺㪼㪼㪻㫀㫅㪾㫊㩷㫆㪽㩷㫋㪿㪼㩷㪉㪇㪇㪋㩷㪠㪜㪜㪜㩷 㪠㫅㫋㪼㫉㫅㪸㫋㫀㫆㫅㪸㫃㩷㪚㫆㫅㪽㪼㫉㪼㫅㪺㪼㩷㫆㫅㩷㪥㪼㫋㫎㫆㫉㫂㫀㫅㪾㪃㩷㪪㪼㫅㫊㫀㫅㪾㩷㩽㩷㪚㫆㫅㫋㫉㫆㫃㩷 㪫㪸㫀㫇㪼㫀㪃㩷㪫㪸㫀㫎㪸㫅㪃㩷㪤㪸㫉㪺㪿㩷㪉㪈㪄㪉㪊㪃㩷㪉㪇㪇㪋㩷
On-demand Data Disseminating with Considering Channel Interference for Efficient Shortest-Route Service on Intelligent Transportation System ∗ Chuan-Ming Liu Comp. Sci. and Info. Eng. National Taipei University of Technology Taipei, Taiwan
[email protected] Abstract – In this paper, we study the shortest route service on Intelligent Transportation System (ITS) which provides the shortest route between two locations specified by a mobile client. We discuss the traditional dedicated channel method and propose a new on-demand broadcasting method. Three measurements are considered when analyzing and simulating the above two methods:(1)the total channel cost (i.e., the amount of data sent from the server), (2) the system time (i.e., the time elapsed between the issuing and termination of a request), and (3) the channel interference. The results show that our on-demand broadcasting method yields less total channel cost for the same group of mobile clients and less system time. Furthermore, as the cell radius increases, the channel interference caused by the broadcast channel increases.
Li-Chun Wang, Lei Chen, Chung-Ju Chang Communication Engineering Department National Chiao-Tung University Hsin-Chu, Taiwan {lichun,tristone,cjchang}@cm.nctu.edu.tw power consumption seriously since the mobile devices can have the power supplied from the vehicles. In this paper, we discuss the shortest-route service on ITS which provides the shortest route information between two locations specified by a mobile client and propose an on-demand broadcasting strategy to fulfill the shortestroute service. To evaluate our scheme, we consider the system time (i.e., the time elapsed between the issuing and termination of a request), the total channel cost (i.e., the amount of data sent from the server), and the channel interference for a group of mobile clients. In the shortest-route service, we observe that different mobile clients may share parts of the route information of each other. Figure 1 shows an example. Traditional client-server model using a dedicated chan-
r0
Keywords: Intelligent Transportation System (ITS), shortest route, on-demand broadcasting, data disseminating, channel cost, system time, and channel interference.
r3
r1 r2
r4 r5
r6 r7
1
Introduction
Intelligent Transportation Systems (ITS) provide many location-aware services, such as map navigation, driving direction, searching nearby hotels or restaurants, and asking weather and traffic information, which can improve safety and efficiency of road transport [4, 5, 7]. For the location-aware services, it is challenging to have mobile clients efficiently receive information in terms of system time and power consumption [1, 8, 9]. However, on ITS, people usually do not consider the ∗ Work
supported by the National Science Concil and MOE for Promoting Academic Excellence of Universities under the grant numbers NSC-90-2213-E-009-068, NSC-92-2213-E-027-010, and EX-91-E-FA06-4-4.
0-7803-8193-9/04/$17.00 ©2004 IEEE
Figure 1: Roads r0 , r3 , r5 , and r6 are for mobile m1 (solid-line) and roads r1 , r3 , r5 , and r7 are for mobile m2 (dashed-line); therefore, roads r3 and r5 are shared by m1 and m2 . nel can be applied to support the shortest-route service but may result in a large channel cost, long system time, large power consumption, and low channel utilization due to repeatedly sending shared information to different mobile clients. Instead of using point-to-point communication, we consider on-demand data broadcasting(/multicasting) instead. Data broadcasting provides an efficient way to disseminate data [1, 2, 6, 9, 13, 14]
701
and can avoid sending the shared information to different mobile clients repeatedly; therefore, yields less total channel cost and shorter system time on average. Since data broadcasting can not be applied to all kinds of the information-centric applications or services [2], it turns out that a new protocol for data broadcasting is necessary. We refer to such broadcasting as on-demand data broadcasting [3]. Our on-demand broadcasting protocol is different from the previous study [2, 3, 10, 12]. In our strategy, the server first collects the requests from mobiles clients, processes the requests in batch, and then broadcasts the result to the mobile clients sent the requests. To avoid waiting for the batch processing, the server collects and processes requests for the next broadcast during the current broadcasting. Besides, since not all the data in the broadcast channel are relevant to the request of a mobile client, we consider that the mobiles can receive results by selective tuning [8, 9, 11, 13, 15], which allows a mobile client to selectively receives the data relevant to its request. A mobile client can tune into the broadcast when the packet is relevant to its request and tune out when the packet is irrelevant. In order that a mobile client can selectively tune into the broadcast, the server needs to set up an index from the data broadcast for mobile clients. Broadcasting data with index has been widely discussed recently [8, 9, 13]. Our proposed on-demand broadcast method using few additional broadcast packets which contain the index. The organization of this paper is as follows. In Section 2, we describe the two models we consider for the shortest-route service, the traditional model and our ondemand model. We analyze these two methods in Section 3. In Section 4, we study the channel interference caused by the broadcast channel and compare it with the one caused by the dedicated channels. The results of the simulation using synthetic data are shown in Section 5. We compare the data broadcasting scheme to the traditional method and measure the performance in terms of the total channel cost and the system time. Besides, we discuss the impact of the data correlation among the routes on the performance of our broadcast scheme for the shortest route services. We then give our conclusion remarks in Section 6.
2
System Models
We assume that there are N mobile clients communicating with a server in the same area. Each mobile client can ask the server for the shortest route information between two specific locations. After receiving the requests, the server processes the requests and will send the corresponding results back to the mobile clients. To send the information back to the mobile clients, the shortest route information forms a number of packets. We represent the packet size as B and the channel capacity between the server and a mobile client
as C. There are two models discussed in this paper: one is the traditional dedicated channel model and the other is the on-demand broadcasting approach.
2.1
Dedicated Channel Method
On the point-to-point client-server system, when a mobile client has a request, it first establishes a dedicated channel with the server. The request is then sent to the server via the dedicated channel. After receiving the request, the server processes the request and then sends the result back to the mobile client via the dedicated channel. By following this approach, when the number of mobile clients is large, a huge channel bandwidth is needed. Since the up-link bandwidth is fixed and limited, many requests may be rejected. Furthermore, a mobile client may waste channel bandwidth while waiting for the server to process the request. In this case, the channel utilization is low and the mobile client consumes unnecessary battery energy.
2.2
On-Demand Broadcasting
To avoid the drawbacks in the dedicated channel method, we propose to use a broadcast channel to serve mobile clients for the shortest route service. We call such an approach as the on-demand broadcasting method. Figure 2 shows an example of an on-demand
broadcast server uplink
Figure 2: On-demand Broadcasting Scenario. broadcasting environment. The server first collects the requests from mobile clients in a period of time and then processes the requests in a batch. We assume that the server has the ability to process a set of the shortest route requests in a batch. After the batch process, the server organizes the results and creates an index. Then the server distributes the results with the index to the mobile clients via a broadcast channel. The index helps the mobile clients to select their own result correctly. In our on-demand broadcasting model for the shortest route service, since the main function of the index is to indicate which packets should be selected for each mobile client, we realize that using a bit array for each mobile client as the index is a simple and efficient way in terms of server workload and total number of packets in the broadcast. The size of a bit array is the number of packets in the broadcast for the results. Each mobile client uses its bit array to select the packets related to
702
its query result. If the packet i is relevant for some mobile client, the ith bit in the bit array for that client is 1; otherwise, the ith bit is 0. Before the server starts broadcasting, it pages all the mobile clients having sent requests and passes the corresponding index (bit-array) to each mobile client. Then, the server starts broadcasting. During the broadcast, the server collects the requests for the next batch and process them to prepare the next broadcast cycle. A diagram for the process in the server is shown in Figure 3.
by the Little law, we get E[N ] = λE[Tsys ]. This implies M ] E[Tsys ] = E[N k=0 k ·pk and E[Ts ] = λ . Since E[N ] = 1 , the expected waiting time is µ M k · pk 1 − . (2) E[Tw ] = k=0 λ µ Form the blocking probability, pb , the total channel cost Tc can be derived as Tc = (1 − Pb ) · M · C,
(3)
where C is the channel capacity. Collecting Requests
Broadcasting
3.2
. .. Time t
Figure 3: A diagram for the process in the server. For a mobile client, it uses a dedicated channel to send a request. After the request has been sent, the dedicated channel is disconnected and can be used by other mobile clients. A mobile client then can do other work or sleep while waiting for the server to tell that the requested result is ready. After receiving the paging signal, a mobile starts to receive the result selectively with the bit-array.
3
Performance Analysis
In this section, we analyze the two approaches in the previous section in terms of the system time Tsys and the channel cost Tc . We assume that the system is in the steady-state.
3.1
Dedicated Channel Method
Let λ and µ be the mobile client arriving rate and the server’s service rate, respectively. We assume that the server has M available channels. If a client sends a request to the server and all the channels are used, the request of this client is blocked. We analyze this scenario. However, in our simulation, to have a fair comparison with the on-demand broadcasting method, we simulate the case when the rejected mobile client re-transmits the request. According to an M/M/m/m queueing model, the blocking probability Pb is Pb =
λ M (µ ) M! M ( µλ )k k=0 k!
.
(1)
We define the total system time Tsys as the summation of the waiting time Tw and the service time Ts ; i.e., Tsys = Tw + Ts . In terms of expectation value,
On-Demand Broadcasting Method
We model the on-demand broadcasting method by the M/G/1 queueing process with service vacations. Refer to Fig. 3 for the server’s process. The mobile clients randomly send requests, the server gives a batch service, and the service ends randomly for each mobile. For the time being, we consider that service time T is fixed. Assume that mobile client i finds that j mobile clients have requests ahead of him and the server serves n mobile clients in each batch. Then, the waiting time wi for the client i is j (4) wi = u(k)ri + T + vi , n where ri is the residual time, vi is the residual vacation time, k is the number of mobiles in the system, nj is the largest integer smaller than nj , and u(k) is the indication function with the following definition 1 k≥1 u(k) = (5) 0 otherwise. Taking the expectation value and applying the Little law, we have λ (6) E[Tw ] = ρE[ri ] + E[Tw ]T + (1 − ρ)E[vi ]. n In our model, we assume the average residual time, vacation time, and service time are T2 . The expected waiting time becomes E[Tw ] =
T /2 , 1 − nλ T
(7)
and the total channel cost for the on-demand broadcasting method is NT . (8) n Obviously, one broadcast channel uses a much smaller bandwidth than M dedicated channels when mobile clients receiving the results. Furthermore, because of being able to disconnect between a mobile client and the server after the request is sent, the server can collect more requests in the on-demand broadcasting model. In the following section, we will discuss the channel interference. Tc =
703
4
Channel Interference
power ratial of on−demand−broadcast−mode and deidicate−channel−mode H=50 70
We assume that there are M dedicated channels for our shortest route service on ITS. For the on-demand broadcasting method, only one channel is used. Intuitively, the interference generated by the dedicated channels is M times the one generated by the broadcast channel. However, the mean transmitted power of the dedicated channels should be less than the mean transmitted power of the broadcasting channel because the base station can transmit less power to maintain the received power for user terminals close to the base station than that for users at the cell boundary. Let Lp be the path-loss and then Lp =
α . r4
(9)
Assume that R denotes the cell radius of a base station. If the mobile clients are uniformly distributed in a cell, the mean path-loss of mobiles in a cell for the dedicated channel method is R 1 α Lpd = · 4 · dr R − d r 0 d0 R 1 α · 4 · dr ≈ R r d0 1 α 1 ·( 3 − = ), (10) R 3d0 3R3 where the value of d0 is much smaller than the value of R, and the mean path-loss for the broadcasting channel is α Lpb = 4 . (11) R Note that we assume there is no mobile client having the distance to the base station smaller than d0 . The relationship of the mean power of these two methods to achieve a fixed SIR target is as follows: Lpb Pdedicated 3H =H· = R 3 . Pon−demand ( d0 ) − 1 Lpd
(12)
From the above equation, we can find that larger cell radius R will lead to higher transmission power for the on-demand broadcasting method. Figure 4 shows this trend. Consequently, how to determine a suitable cell radius for the on-demand-broadcast method becomes important. If the radius is too large, the transmission power becomes much larger than the transmission power of the dedicate channels. In that case, the interference caused by the broadcasting channel is larger than that caused by the dedicated channels.
5
Numerical Results
We present our simulation work on the system time, Tsys , and the total channel cost, Tc . The data set is synthetic and consists of line segments on the plane. The
60
power ratial (dB)
50
40
30
20
10
0 1000
d0=100 d0=50 d0=10 2000
3000
4000
5000 6000 Cell radius (meters)
7000
8000
9000
10000
Figure 4: The power ratio of the on-demand broadcasting method to the traditional dedicated channel method using H=50 dedicated channels shortest route information for a mobile client is a subset of these line segments. In our simulation, we first generate 10000 line segments uniformly in the unit square as the data set and then let the request of a mobile client be a subset of these line segments. We consider 1000 mobile clients. Since different mobile clients can share parts of the route information, a line segment is randomly referred many times by different mobiles. We control the data correlation by setting a percentage of reference to each line segment. According to the ondemand broadcasting method, the server generates an index (bit-array) for each mobile client in each batch and a mobile client uses the index to receive the information. Each packet is 24 bytes containing a line segment. The data rate of a channel is 288 kbps. The mobile client arrival rate is 50 arrivals per second. Our numerical results are tailored toward the performance analysis in Section 3 for the steady-state. In our simulation, we consider that the number of dedicated channels, M , is equal to 20, 30, 40, and 50 for the dedicated channel method and the maximum number of served clients for the on-demand method is 20, 30, 40, and 50. The backoff time is 0.1 second and the data correlation between the routes is from 40% to 90%. The performance metrics are the system time Tsys and the total channel cost Tc . Figure 5 shows the mean system time for different data correlations. For the dedicated channel method, as the data correlation increases, the total amount of packets for sending the route information to each mobile client increases and thereby results longer system time. As the number of channels becomes larger, the waiting time is reduced and leads to a shorter average system time. If the waiting time is very small, for the case of 50 channels, the mean system time is almost the same as the the mean service time no matter what the data correlation is. For the on-demand method, the broad-
704
35
Dedicated channel method 100 80
number of channel=50 number of channel=40 number of channel=30 number of channel=20
25
60 40 20 0 0.4
Dedicated method:30 channels On−demand method:max.# =40
30
mean system time(second)
mean system time(second)
120
0.45
0.5
0.55
0.6
0.65 0.7 data correlation
0.75
0.8
0.85
0.9
20
15
10
Different max.# of served clients per batch (on−demand method) 5
max.# =50 max.# =40 max.# =30 max.# =20
150
0 0.4
0.45
0.5
0.55
0.6
100
0 0.4
0.65 0.7 data correlation
0.75
0.8
0.85
0.9
0.85
0.9
(a) Mean service time
50
1
0.95
0.45
0.5
0.55
0.6
0.65 0.7 data correlation
0.75
0.8
0.85
0.9 0.9
mean service time (second)
mean system time(second)
200
Figure 5: The mean system time for the dedicated channel method and the on-demand method.
0.85
0.8
0.75
0.7
0.65
0.6
Dedicated method:30 channels On−demand method:max.# =40 0.45
0.5
0.55
0.6
0.65 0.7 data correlation
0.75
0.8
(b) Figure 6: The system time (a) and service time (b) for the dedicated channel method with 30 channels and the on-demand method with the maximum number of served clients 40. data correlation is since it depends on the distribution of the results of the requests rather than the amount of requested packets by a client. Dedicated channel method
6
total channel cost(bit)
10
x 10
number of channel=50 number of channel=40 number of channel=30 number of channel=20
9.5 9 8.5 8 7.5 7 0.4
0.45
0.5
0.55
6
0.6
0.65 0.7 data correlation
0.75
0.8
0.85
0.9
On−demand method
5
total channel cost(bit)
cast packets can be shared. Thus, the mean service time for different data correlations is the same and we will illustrate this phenomenon later. For a fixed maximum number of served clients, the mean waiting time for different data correlation is also the same since the waiting time is independent from the data correlation in the on-demand method. As the maximum number of served clients becomes larger, the probability for a mobile client to wait for the service is less and hence the system time decreases. In our on-demand broadcasting method, a mobile client sends a request and then disconnects the channel and waits for the result. During the waiting period, the channel can be used by other mobile clients. It is reasonable to assume that the maximum number of mobile clients served in a batch is larger than the number of available channels in the dedicated channel method. We compare the result of the dedicated channel method with 30 channels and the result of the ondemand method with the maximum number of served mobile clients 40 as in Fig. 6(a). The comparison shows that when the data correlation is high, our on-demand method yields a better system time due to the ability to share the packets between mobile clients. This result suggests what the maximum number of served clients should be adopted in a batch to have a better system when using the on-demand method. Since there is no extra indexing packets in the dedicated channel method, the service time of the dedicated channel method is better than that of the on-demand method. Our simulation shows that the service time is much smaller than the waiting time as comparing in Fig. 6. The system time is hence dominated by the waiting time for both methods. The service time and waiting time for the dedicated channel method increase as the data correlation increases since each mobile client needs more packets. The mean service time for the ondemand method is almost the same no matter what the
0.55 0.4
x 10
max.# =50 max.# =40 max.# =30 max.# =20
5
4
3
2 0.4
0.45
0.5
0.55
0.6
0.65 0.7 data correlation
0.75
0.8
0.85
0.9
Figure 7: The channel cost for the dedicated channel method and the on-demand method. Figure 7 shows the total channel costs of the two methods. The total channel cost indicates the total amount of data sent by a server to serve a group of mobile clients. For the dedicated channel method, the total channel cost increases as the data correlation becomes higher. On the other hand, the total channel cost does not change in our on-demand broadcasting method for different data correlations. The conclusion is obvious since our on-demand broadcasting method
705
always sends almost the same amount of data regardless of the data correlation. If the maximum number of served clients decreases in each batch, the server needs more batches to server the mobile clients. Hence, the total channel cost increases as the maximum number of served clients decreases. By contrast, for the dedicated channel method, as the data correlation increases, the amount of data to be sent increases. Note that the total amount of data to be sent is fixed for the server since the server sends the corresponding result to each mobile client independently. So, the result also shows that the total channel cost is the same for different numbers of channels. Figure 8 shows the ratio of the total channel cost of the dedicated channel method to that of the ondemand method. The result shows that our on-demand # of channels v.s. max.# of served clients per batch 45
40
#=50 #=40 #=30 #=20
ratio of total channel cost
[1] S. Acharya, M. Franklin, and S. Zdonik. Balancing push and pull for data broadcasts. In Proceedings of the 1997 ACM SIGMOD International Conference on Management of Data, pages 183–194, May 1997. [2] S. Acharya and S. Muthukrishnan. Scheduling on-demand broadcasts: New metrics and algorithms. In Proceedings of the 4th Annual ACM/IEEE International Conference on Mobile Computing and Networking (MobibCom ’98), pages 43–54, Dalla, TX, October 1998.
[4] O. Andrisano, R. Verdone, and M. Nakagawa. Intelligent transportation systems: the role of third- generation mobile radio network. IEEE Communication Magazine, pages 144– 151, 2000.
30
25
[5] K. Biesecker. Broadband wireless, integrated services, and their application to intelligent transportation systems. http://www.its.dot.org.
20
15
0.45
0.5
0.55
0.6
0.65 0.7 data correlation
0.75
0.8
0.85
0.9
Figure 8: The ratio of the total channel cost of the dedicated channel method to the total channel cost of the on-demand method. broadcasting method allows the server to send much less amount of data to the same group of mobile clients with respect to the amount of data sent by the dedicated channel method. Consider the case of 50 channels in the dedicated channel method and 50 served clients in the on-demand broadcasting method. The ratio of total channel cost is at least 30. For the on-demand method, we use one broadcast channel to server as many clients as possible. However, the number of clients served by a channel in the dedicated channel method is limited.
6
References
[3] D. Aksoy, M. J. Franklin, and S. Zdonik. Data staging for on-demand broadcast. In Proceedings of the 27th Very Large Data Bases Conference, pages 571–580, 2001.
35
10 0.4
ulation also shows that the performance is significantly improved when the data correlation among the routes is high. The simulation also suggests a way to adopt the maximum number of served clients in a batch when applying our on-demand broadcasting method. Furthermore, we explore the channel interference caused by the broadcasting channel. It shows that the channel interference increases when the cell radius increases and suggests a way to decide a suitable cell radius.
Conclusions
In this paper, we consider the shortest route service on ITS. We provide an on-demand broadcasting method in which the server collects the shortest route requests from the mobile clients, processes the requests in a batch, and sends the results by broadcasting the results with an index. The mobile clients receive their shortest route information by selectively tuning into the broadcast. Our on-demand broadcasting method can provide less total channel cost and fewer system time. Our sim-
[6] A. Datta, D. E. VanderMeer, A. Celik, and V. Kumar. Broadcast protocols to support efficient retrieval from databases by mobile users. ACM Transactions on Database Systems, 24(1):1–79, March 1999. [7] K. Furukawa. Status and future direction of intelligent drive assist technology. In Proceedings on IEEE Intelligent Transportation Systems, pages 1000–1005, 2001. [8] S. Hambrusch, C.-M. Liu, W. Aref, and S. Prahakar. Query processing in broadcasted spatial index trees. In Advances in Spatial and Temporal Databases - 7th International Symposium (SSTD 2001), Lecture Notes in Computer Science 2121, pages 502–521. Springer-Verlag, July 2001. [9] T. Imieli´ nski, S. Viswanathan, and B. R. Badrinath. Data on air: Organization and access. IEEE Transactions on Knowledge and Data Engineering, 9(3):353–372, May/June 1997. [10] W.-C. Lee, Q. Hu, and D. L. Lee. A study on channel allocation for data dissemination in mobile computing environments. Mobile Networks and Applications, 4(2):117–129, 1999. [11] N. Shivakumar and S. Venkatasubramanian. Efficient indexing for broadcast based wireless systems. Mobile Networks and Applications, 1(4):433–446, May/June 1996. [12] W. Sun, W. Shi, B. Shi, and Y. Yu. A cost-efficient scheduling algorithm of on-demand broadcasts. Wireless Networks, 9(3):239–247, 2003. [13] K.L. Tan, J. X. Yu, and P.K. Eng. Supporting range queries in a wireless environment with nonuniform broadcast. Data Knowledge Engineering, 29(2):201–221, 1999. [14] N. Vaidya and S. Hameed. Scheduling data broadcast in asymmetric communication environments. ACM/Baltzer Journal of Wireless Network, 5(3):171–182, 1999. [15] J.X. Yu and K.L. Tan. An analysis of selective tuning schemes for nonuniform broadcast. Data and Knowledge Engineering, 22(3):319–344, 1997.
706
