三、 Approximate Location Management
3.5 Extension
In section 3, we propose the dynamics servers of location management system in wireless sensor networks. We use the history of message flow to adapt the location of home server and introduce how the snapshot query works. The snapshot query is used to query the location of target at specific time. Since a history of message flow is the main factor of dynamic server, the query distribution of next period is more similar with history will improve the performance of our work. The contrary of snapshot query is continuous query. Continuous query could seen as a fixed query distribution of snapshot queries. A continuous query message has to append the end time and the rate requirement. Users send continuous queries to get the replies from server in a specific rate. This large replies flow encourages us to take continuous query into consideration in our work.
We illustrate how the continuous query works in basic system first. The basic system is a location service without dynamic server mentioned before. We add a new table to deal with continuous query. If home server receives the approximate query and the information in home server can satisfy, it adds an entry in this table and reply the answer in a fixed period. Otherwise, this query message is forwarded to local server and local server will be responsible to reply. Since the local server of a target is not fixed, the hand off problem will be exist. When a target moves for a long time and selects another node to be a new local server, the old local server has to send the entry about target form table to new local server. Therefore, the local update message has to append the old local information to inform new local server. With dynamic server, home server could change its location. So we use same method to solve the hand off problem.
The main modification of our method for continuous query is add more count when node replies. According to the information of continuous query, the server could know the query count in the following period. So we can add the expectable query count directly without receiving a query. However, if we add all count once at one node, the server will only move or split once. The server has to add these count and bring it to next splitting or moving node to add. Therefore, if a user sends a continuous query with a continuous rate rc and end time tend, the server node adds the count when it replies and the quantity is tcheck× rc× tend when it checks the neighbor table.
Another extension is load balance. The hot spot problem could cause a small portion of nodes out of energy. In GPSR, each node sends beacon message to maintain the accuracy of neighbor table. We append residual energy to this beacon message to let each node know the residual energy of neighbors. Before a node decides to move or split the server to neighbor, it has to ensure that the energy of neighbor is larger than the threshold which user define. When the node detects that its residual energy is less than the threshold, it sends moving message to the node which has most energy from its neighbor nodes. Certainly, the energy of this neighbor node has to be larger than the threshold.
Chapter 4
Performance Evaluation
4.1 Simulation Model
We developed a simulator based on ns-2(version 2.29)[18] to compare approximate location man-agement with dynamic server scheme to EASE[13]. To use GPSR protocol in our simulation, we also use the routing protocol in HLS(Hierarchical Location Service for Mobile Ad-hoc Networks) patch for ns-2.29[19]. We deploy 100 sensor nodes on a 2000 × 2000m2 field. The field is divided into 200 × 200m2 grid cells. There is one sensor node in the center of each cell. The MAC protocol in our simulator is based on IEEE 802.11 and the transmission range of each node is 250 m. Table 4.1 summarizes the system parameters and setting.
The mobility model in our simulator is linear model. The setting of moving target includes the destination, the start moving time, and the speed. We change the destination every ten seconds to avoid that the target has arrived the destination and be static. The speed of target influences the count of remote update. When target moves faster, the detecting node has to send more remote update message to home server. Therefore, we adjust the speed of target to control the remote update rate. The locations of querying nodes for sending approximate query are issued randomly from the sensor nodes in the field. The error bound of queries are uniformly distributed between 0 and 300 m.
Parameter Setting
Field Size 2000 × 2000m2
Number of Nodes 110
Radio Range 250 m
Sensor Sampling Rate 2 sec
Record Reduce Rate (α) 0.5
Checking Period (tcheck) 2 sec
Query Rate 1 - 10 /sec
Error Bound of Query 0 - 300 m
Speed of Target 10 - 50 m/sec
Approximate Radius 100 m
Initial Energy 2J
Eelec 50nJ
²f s 10pJ/bit/m2
²ap 0.0013pJ/bit/m4
Threshold Distance(d0) 75 m
Update Message Payload Size 32 bytes Reply Message Payload Size 32 bytes Acknowledge Message Payload Size 16 bytes
Query Start Time 10 sec
Simulation Time 500 sec
Table 4.1: Simulation parameters
4.2 Simulation Result
4.2.1 Message Size
This section evaluates the message size of ALM. We first take the query rate into consideration.
Both the merge threshold and split threshold are 3 and the move threshold is set to 6. The speed of target is 30 m/sec. The result is shown in Figure 4.1. The performance of ALM is always better than EASE. The higher query rate can decrease the total message size of ALM obviously and the message size of EASE is in proportion to the query rate. This is because that the lower query rate could waste the move and split message since the benefit from query message is lower.
Therefore, our work is suitable for higher query rate.
Figure 4.1 shows the impact on total message size of target’s moving speed. The settings of parameter are the same besides moving speed. The query rate is 4/sec. We can see that the moving speed has a little effect on the performance of EASE and ALM. No matter how fast the
0
(a) Message size vs. query rate
0
(b) Message size vs. moving speed Figure 4.1: Message size Figure 4.2: Message size vs moving threshold
speed is, the message size of ALM is always less than EASE.
We now evaluate the moving threshold of total message size. At first, we use random query which means the querying sources are randomly from the sensor node in the field as shown in Figure 4.2. To observe the moving situation easily, we set the higher splitting threshold. The splitting and merge threshold are set to 10. The server of EASE does not move or split, so the moving threshold cannot influence the message size. However, the performance of ALM is not stable but it is still better than the EASE. The lower moving threshold causes the server to move more easily. It means that moving server frequently under random query could not get more benefit. The trend of query flow is not apparent when querying source is random. Then we let querying sources are bias. The sensing field is divided into 10 × 10 grid as we mentioned before.
Now, we let all query from the 3 × 3 grid at lower right corner of field. The result is in Figure 4.2. The performance is the best when the moving threshold is 6. Because the lower threshold
0
Figure 4.3: Check period vs message size
0
Max error bound / apprximate radius
Message size (k bytes)
EASE ALM
Figure 4.4: Error bound vs message size
makes the ping-pong effect. The server moves into the little grid which the querying sources quickly when moving threshold is lower. Then it moves from one node to another one. After a while, it moves back again and waste the message. However, the higher threshold does not promise the better performance. When the threshold is higher, the server moves slowly and the query message has to route a long distance for a while.
Figure 4.3 shows the performance of ALM to check period. We set the check period from 1 sec to 5 sec. When the check period is 1 sec, the message size is obviously higher. It is because that the check period is too short and the record of location server is not enough to decide if it should move or split.
Both the count of remote update message and forwarding query message are affected by the approximate radius. In Figure 4.4, we compare the performance of ALM and EASE when we set different ratio of error bound and approximate radius. The query rate is 2/sec. We can see the message size of EASE is higher when the ratio is smaller. This is because the query messages
0
request reply update update ack
Message count
EASE ALM
Figure 4.5: Breakdown of message
are forwarded when the error bound is smaller than the approximate radius. The smaller ratio makes more query forward to local server and decrease the count of remote update. Therefore, the message size is less when the ratio grows. The slop of ALM is similar to EASE. The reason is the same as mentioned above, but the performance of ALM is better.
To know how ALM improves the performance, we provide a breakdown of message in Figure 4.5. The update message includes the local update and forwarding update. Forwarding update messages are the overhead of replica server nodes. In ALM, it decrease the most part of message in query and reply and incurs a little overhead.
4.2.2 Quality of Query
We also evaluate the quality of query in ALM scheme and EASE scheme. We divide the quality of query into two parts, success rate and latency. The success rate means the percentage of receiving replies. Users could be worry about if they can receive a reply when they query to dynamic servers. To ensure that querying node can receive a reply in ALM scheme, we evaluate the success rate of query rate. The latency includes the total time from sending query to receiving the answer. We expect the latency is lower in ALM scheme since the servers should move or split closely to querying node when query rate is higher. The results are in Figure 4.6.
As shown in Figure 4.6, when the query rate is higher, both the latencies of ALM and EASE are lower. But the performance of ALM is always better than EASE especially in high query rate. The higher query rate encourages the servers to split or move and the routing distance
0 Figure 4.6: Quality of query
from querying node to dynamic servers is shortened. Then the latency is decreased. The success rate of query represents the reliability of location service. Therefore, we evaluate the success rate of different query rate. The result is shown in right figure of Figure 4.6. Both EASE and ALM has high reliability. The success rate is always over 98%.
4.2.3 Energy Consumption
We now evaluate the energy consumption in this subsection. The energy consumption model and the parameters refer to [15]. First measurement is minimum energy. It means the minimum residual energy of node in the network field. This measurement represents the load balance since the higher minimum residual energy means each node costs less power. As shown in left figure of Figure 4.7, the performance of EASE decreases soon when query rate increases. Because the server of EASE cannot move or split, the nodes surrounding server are routed through a lot of message. In ALM scheme, the performance decrease slightly when query rate increases and the residual energy is more. The splitting servers share responsibility for serving queries, so ALM could balance the load. The second measurement is total energy consumption. We also increase the query rate to observe the effect of energy consumption. The result is shown in the right figure of Figure 4.7. The energy consumption of ALM is always less than EASE since the message count is less that we mentioned before.
0
(a) Minimum energy vs. query rate
0
(b) Total energy vs. query rate Figure 4.7: Energy consumption
Percentage of continuous queries (%)
Message count
Figure 4.8: Message count vs. percentage of continuous queries
4.2.4 Others
In section 3.5, we illustrate how the continuous query works in our scheme. Now we evaluate its performance in terms of message count and minimum energy to percentage of continuous query. The simulation time is changed to 200 sec. The query rate is set to 1 and each continuous query rate is set randomly distributed between 1 to 4. The continuous query duration is issued randomly between 0 seconds and 100 second. In Figure 4.8(a), with percentage of continuous query is increasing, the ratio of ALM to EASE is decreasing. It means that the improvement of our scheme is better when the amount of continuous queries is more. The load balance is still good in our scheme as show in Figure 4.8(b).
To show the ALM scheme can adapt the message flow, we use two kinds of query distribution to evaluate the message count in each time period. The query distribution is random. The result is shown in Figure 4.9. When simulation just starts, the message count of ALM is the same
0 100 200 300 400 500 600 700 800 900
0 10 20 30 40 50 60 70 80 90 100
Time(sec)
Message count
EASE ALM
Figure 4.9: Message size vs. time series
as EASE. Then the message count of ALM decreases soon after twenty seconds. It means the dynamic servers adjust their positions according to the message flow and reduce the message count.
Chapter 5 Conclusions
In this thesis, we proposed an approximate location management scheme for object tracking in sensor networks. This scheme consists of approximate queries and dynamic servers. The approximate queries to let target’s information be stored in local storage to reduce the update message cost. The dynamic servers utilize the history of message flow to adjust the location of centric storage. We let server move to where has more message and split a replica to serve high query rate region. It reduce total message count of network field and conserve the energy consumption of sensor nodes. The replica nodes also balance the load of servers. We have evaluated the performance of the proposed location management with simulation. The simulation results show that the ALM scheme could improve the performance and outperform the EASE scheme especially in the high query rate.
Bibliography
[1] C. Gui and P. Mohapatra. Power conservation and quality of surveillance in target tracking sensor networks. In Proceedings of the 10th Annual International Conference on Mobile Computing and Networking, pages 129–143, 2004.
[2] C. Intanagonwiwat, R. Govindan, and D. Estrin. Directed diffusion: A scalable and robust communication paradigm for sensor networks. In Proceedings of the ACM/IEEE Interna-tional Conference on Mobile Computing and Networking, pages 56–67, 2000.
[3] X. Ji and H. Zha. Sensor positioning in wireless ad-hoc sensor networks using multidi-mensional scaling. In Proceedings of Twenty-third AnnualJoint Conference of the IEEE Computer and Communications Societies, 4, 2004.
[4] B. Karp and HT Kung. GPSR: Greedy Perimeter Stateless Routing for wireless networks. In Proceedings of 6th Annual International Conference on Mobile Computing and Networking, pages 243–254, 2000.
[5] HT Kung and D. Vlah. Efficient location tracking using sensor networks. In Proceddings of IEEE Wireless Communications and Networking, 3, 2003.
[6] J. Li, J. Jannotti, D.S.J. De Couto, D.R. Karger, and R. Morris. Scalable location service for geographic ad hoc routing. In Proceedings of ACM 6th Annual International Conference on Mobile Computing and Networking, pages 120–130, 2000.
[7] J. Li, Y. Pan, and X. Jia. Analysis of dynamic location management for PCS networks.
IEEE Transactions on Vehicular Technology, 51(5):1109–1119, 2002.
[8] C.Y. Lin, W.C. Peng, and Y.C. Tseng. Efficient In-Network Moving Object Tracking in Wireless Sensor Networks. IEEE Transactions on Mobile Computing, 5(8):1044–1056, 2006.
[9] A. Mainwaring, D. Culler, J. Polastre, R. Szewczyk, and J. Anderson. Wireless sensor networks for habitat monitoring. In Proceedings of the 1st ACM international workshop on Wireless sensor networks and applications, pages 88–97, 2002.
[10] S. Ratnasamy, B. Karp, L. Yin, F. Yu, D. Estrin, R. Govindan, and S. Shenker. GHT: a geographic hash table for data-centric storage. In Proceedings of the 1st ACM international workshop on Wireless sensor networks and applications, pages 78–87, 2002.
[11] S. Tabbane. Location management methods for third generation mobile systems. Commu-nications Magazine, IEEE, 35(8):72–78, 1997.
[12] X. Wu. VPDS: Virtual Home Region Based Distributed Position Service in Mobile Ad Hoc Networks. In Proceedings of 25th IEEE International Conference on Distributed Computing Systems, pages 113–122, 2005.
[13] J. Xu, X. Tang, and W.C. Lee. EASE: An energy-efficient innetwork storage scheme for object tracking in sensor networks. In Proceedings of IEEE Communications Society Con-ference on Sensor and Ad Hoc Communications and Networks, 2005.
[14] Y. Xue, B. Li, and K. Nahrstedt. A scalable location management scheme in mobile ad-hoc networks. In Proceedings of IEEE Conference on Local Computer Networks, 2001.
[15] O. Younis and S. Fahmy. Distributed Clustering in Ad-hoc Sensor Networks: A Hybrid, Energy-Efficient Approach. In Proceedings of Twenty-third AnnualJoint Conference of the IEEE Computer and Communications Societies, 1:629–640, 2004.
[16] W. Zhang and G. Cao. DCTC: dynamic convoy tree-based collaboration for target tracking in sensor networks. IEEE Transactions on Wireless Communications, 3(5):1689–1701, 2004.
[17] W. Zhang and G. Cao. Optimizing tree reconfiguration for mobile target tracking in sensor networks. In Proceedings of Twenty-third AnnualJoint Conference of the IEEE Computer and Communications Societies, 4, 2004.
[18] The network simulator - ns-2. http://www.isi.edu/nsnam/ns/.
[19] HLS patch for ns-2.29. http://www.cn.uni-duesseldorf.de/staff/kiess/software/hls-ns2-patch.