Experiment Results

Figure 12: Experiment 1: Impact of scalability

4.3 Experiment Results

Experiment 1: Impact of Scalability

In this section, we use simulation to compare MobiServNet against flooding. Although we have examined almost all the combinations of factors described above, we discuss only the most salient ones here, for lack of space. As shown in Figure 12, the left diagram plots the aver-age advertisement costs for MobiServNet (for flooding, of course, the insertion costs are zero).

MobiServNet incurs less per event overhead in registering advertisements. The right diagram plots the average query cost for flooding, the bounded uniform query size distribution, and the exponential distribution. For this graph we use a uniform advertisement distribution, since the advertisement distribution does not aect the query delivery cost. For this simulation, the bound of the bounded uniform distribution was ¹₄wk the size of the largest possible query.

MobiServNet-U is MobiServNet with uniform advertisement distribution. MobiServNet-BU and MobiServNet-E mean MobiServNet with bounded uniform and exponential query range distribution respectively. Even for this generous query range size, MobiServNet perform quite well (almost a third the cost of flooding when N=300). The average query size for the

ex-ponential distribution was set to be ₁₆¹ wk the largest possible query. The superior scaling of MobiServNet is evident in these graphs. Clearly, this is the vision that might expect MobiS-ervNet to perform best, when most of the queries are small and relatively rare queries are large.

Experiment 2: Impact of Update

In this experiment, we study the performance under dierent average speed of all objects y.

When the temporal and spatial groups are formed, the group representative will cache the information of recently member positions. The group member does not perform recently update to the responsible directory if streams of updates still belong to the same group range (we referred as a group hit). In fact, streams of updates will belong to the same group with high probability. But the simulation use an approach based on the random waypoint mobility model[15]. Surely, the implementation of the latter one is more easier than the former one.

Mobile object update their advertisements with positions every 125 milliseconds. A posi-tion consists of two attributes: an x-coordinate and a y-coordinate. This simulaposi-tion runs for 300 seconds. The positions of mobile objects vary within [0,1]²⁴ virtual space. As shown in the Figure 13, we see that this update mechanism significantly keeps good hit ratio within the groups (by up to 95% with 12,000 nodes). In the Figure 14, MobiServNet-R means Mo-biServNet whose updates are produced when positions are changed. MoMo-biServNet-G means MobiServNet whose updates are produced when group memberships are changed or position of the group representative changed. Significantly, we see that this group update mechanism reduces the total number of updates to the responsible directories (by up to 45% with 12,000 nodes, compare MobiServNet-G to MobiServNet-R).

Experiment 3: Impact of Directory Storage

While increasing the number of mobile objects from 2,000 to 12,000 (varies number of directories from 50 to 300), we measure the storage size of the 3-d R-tree and binary Patricia

˃

˄˃

˅˃

ˆ˃

ˇ˃

ˈ˃

ˉ˃

ˊ˃

ˋ˃

ˌ˃

˄˃˃

˃ˁ˄˅ˈ ˃ˁˈ ˅ ˋ ˆ˅

˴̉˸̅˴˺˸ʳ̆̃˸˸˷ʳ̂˹ʳ˴˿˿ʳ̂˵˽˸˶̇̆

˻˼̇ʳ̅˴̇˼̂

˅˃˃˃ʳ̂˵˽˸˶̇̆

˄˅˃˃˃ʳ̂˵˽˸˶̇̆

Figure 13: Experiment 2: Impact of update

˃ ˈ˃˃˃˃

˄˃˃˃˃˃

˄ˈ˃˃˃˃

˅˃˃˃˃˃

˅ˈ˃˃˃˃

˃ˁ˄˅ˈ ˃ˁˈ ˅ ˋ ˆ˅

˴̉˸̅˴˺˸ʳ̆̃˸˸˷ʳ̂˹ʳ˴˿˿ʳ̂˵˽˸˶̇̆

ʶʳ̂˹ʳ̈̃˷˴̇˸̆

ˠ̂˵˼˦˸̅̉ˡ˸̇ˀ˥

ˠ̂˵˼˦˸̅̉ˡ˸̇ˀ˚

Figure 14: Experiment 2: Impact of update.

˃ ˈ˃˃˃˃

˄˃˃˃˃˃

˄ˈ˃˃˃˃

˅˃˃˃˃˃

˅ˈ˃˃˃˃

ˆ˃˃˃˃˃

ˆˈ˃˃˃˃

ˇ˃˃˃˃˃

˅˃˃˃ ˇ˃˃˃ ˉ˃˃˃ ˋ˃˃˃ ˄˃˃˃˃ ˄˅˃˃˃

ʶʳ̂˹ʳ̀̂˵˼˿˸ʳ̂˵˽˸˶̇̆

̆̇̂̅˴˺˸ʳ̆˼̍˸ʳʻ˞˕ʼ

ˆ˷ʳ˥ˀ̇̅˸˸

˵˼́˴̅̌ʳˣ˴̇̅˼˶˼˴ʳ̇̅˼˸̆

Figure 15: Experiment 3: Impact of directory storage

tries. As shown in Figure 15, the binary Patricia tries (BPT) which stores the locations in binary string representation consumes about 56% of the storage space spent by the 3-d R-tree.

Therefore, the reduction ratio of storage space was approximately 44%.

Experiment 4: Impact of Retrieval Cost

Figure 16 plots the number of query results received for dierent query range sizes. It appears that we received fewer results than expected since the reason we have explained in the analysis. MobiServNet-NK is MobiServNet without keyword constraints, and another one with keyword constraints.

5 Conclusions

In this paper, we devised an eective service infrastructure for discovering mobile services.

Our location data model captures mobile objects as binary strings in one dimensional space instead of conventional (x, y) geographic coordinates in two dimensional space that are in-dexed by R-tree relevance techniques, and thus can reduce storage cost while improving query processing. Our temporal and spatial grouping method provides a better update mechanism

˃

Figure 16: Impact of retreival cost

and improves the query performance. Our appropriate approximated query results provides the query performance better than all possible results. The design of MobiServNet is fully distributed and more scalable. The simulation studies confirm all above declarations. We also have planed to conduct larger scale experiments in the future.

References

[1] T. Asano, D. Ranjan, T. Roos, E. Welzl, and P. Widmayer. Space-filling curves and their use in the design of geometric data structures. Theoretical Computer Science, 181(1):3—15, 1997.

[2] N. Beckmann, H.-P. Kriegel, R. Schneider, and B. Seeger. The r*-tree: an e!cient and robust access method for points and rectangles. SIGMOD Rec., 19(2):322—331, 1990.

[3] J. L. Bentley and J. H. Friedman. Data structures for range searching. ACM Comput.

Surv., 11(4):397—409, 1979.

[4] A. Guttman. R-trees: a dynamic index structure for spatial searching. In SIGMOD ’84:

Proceedings of the 1984 ACM SIGMOD international conference on Management of data, pages 47—57, New York, NY, USA, 1984. ACM Press.

[5] M. Harren, J. Hellerstein, R. Huebsch, B. Loo, S. Shenker, and I. Stoica. Complex queries in dht-based peer-to-peer networks, 2002.

[6] G. Kollios, D. Gunopulos, and V. J. Tsotras. On indexing mobile objects. In PODS, pages 261—272. ACM Press, 1999.

[7] H. P. Kriegel, M. Schiwietz, R. Schneider, and B. Seeger. Performance comparison of point and spatial access methods. In SSD ’90: Proceedings of the first symposium on Design and implementation of large spatial databases, pages 89—114, New York, NY, USA, 1990.

Springer-Verlag New York, Inc.

[8] D. Pfoser. Indexing the trajectories of moving objects. IEEE Data Eng. Bull., 25(2):3—9, 2002.

[9] S. Ratnasamy, P. Francis, M. Handley, R. M. Karp, and S. Shenker. A scalable content-addressable network. In SIGCOMM, pages 161—172, 2001.

[10] A. Rowstron and P. Druschel. Pastry: Scalable, decentralized object location, and routing for large-scale peer-to-peer systems. Lecture Notes in Computer Science, 2218:329—??, 2001.

[11] S. Saltenis, C. S. Jensen, S. T. Leutenegger, and M. A. Lopez. Indexing the positions of continuously moving objects. In W. Chen, J. F. Naughton, and P. A. Bernstein, editors, SIGMOD Conference, pages 331—342. ACM, 2000.

[12] H. Samet. The design and analysis of spatial data structures. Addison-Wesley Longman Publishing Co., Inc., Boston, MA, USA, 1990.

[13] B. Seeger and H.-P. Kriegel. Techniques for design and implementation of e!cient spatial access methods. In VLDB ’88: Proceedings of the 14th International Conference on Very Large Data Bases, pages 360—371, San Francisco, CA, USA, 1988. Morgan Kaufmann Publishers Inc.

[14] I. Stoica, R. Morris, D. R. Karger, M. F. Kaashoek, and H. Balakrishnan. Chord: A scalable peer-to-peer lookup service for internet applications. In SIGCOMM, pages 149—

160, 2001.

[15] K. Wang and B. Li. Group mobility and partition prediction in wireless ad-hoc ntworks, 2002.