Impact of Client’s Maximum Step

In the first experiment, we observe the performance of our algorithm by varying the maximum moving step of clients. In the simulation model, the clients will access several data after moving a step. With the increase of the clients’ moving step, the access records will increase too. Both the client and the base station could gather more statistic for further cache replacement. The simulation results are shown in Figure 9, Figure 10, Figure 11, and Figure 12. As shown in Figure 9, we can see that the client hit ratio of CCR outperforms other algorithms. The average improvement of client hit ratio over the LRU and LFU is about 50%. The performance of LRU is the worst and it is not influenced by the variation of clients’ moving step. It is because that the LRU just takes the access time into consideration. The LFU performs better

0

Figure 9: Client hit ratio under various of the client’s maximum moving step

10

Base station hit ratio (%)

CCR LRU LFU

Figure 10: Base station hit ratio under various of the client’s maximum moving step than LRU since that LFU can collect more statistic of access records with the increase of the client’s moving step. The same as CCR, if the clients move further, the CCR can gather more information of the access patterns and popularity of objects in the environment to do cache replacement precisely. In Figure 10, we can observe that all three algorithms perform well when the clients move further, because that all the base stations almost store the objects which are popular among clients. In Figure 11 and Figure 12, the query cost of CCR is much smaller than those of LRU and LFU since the cache hit ratio of CCR is higher.

150

Figure 11: Client query cost under various of the client’s maximum moving step

30

Figure 12: Base station query cost under various of the client’s maximum moving step

0

Figure 13: Client hit ratio under various cache capacity 4.2.2 Impact of Client Cache Capacity

In the second experiment, we investigate the influence of the various client cache capacities.

The simulation results are shown in Figure 13, Figure 14, Figure 15, and Figure 16. Of course the hit ratio will be improved while the cache size is relative large. But as Figure 13, it shows that the CCR can use the cache storage more efficiently than LRU and LFU. When the client’s cache size is 12% to the base station, the improvements of client hit ratio over the LRU and LFU are 72% and 35% respectively. In Figure 14, the base station hit ratio gets smaller because in the same time the client hit ratio is getting larger, but the CCR still performs much better than others. Figure 15 and Figure 16 show the query costs of clients and base stations.

4.2.3 Impact of Maximum Object Size

Then we observe the performance under various maximum sizes of objects. The simulation results are shown in Figure 17, Figure 18, Figure 19, and Figure 20. If the possible maximum size of objects is large, the cache insufficiency will happen frequently. If the cache space is occupied by large size objects, the number of cached objects will be small, so that the hit ratio will be reduced too. The initial value of client cache is 250, so as Figure 17, all three algorithms

20

Base station hit ratio (%)

CCR LRU LFU

Figure 14: Base station hit ratio under various cache capacity

170

Figure 15: Client query cost under various cache capacity

32

Figure 16: Base station query cost under various cache capacity

0

Figure 17: Client hit raio under various maximum sizes of objects

20

Base station hit ratio (%)

CCR LRU LFU

Figure 18: Base station hit raio under various maximum sizes of objects

perform badly when the maximum size of object is approximate to 150, and it reaches 60%

of client cache. But in other cases when the maximum sizes of objects are relatively small, the CCR performs much better than other algorithms. In Figure 18, as the maximum size of object gets bigger and the client cache hit ratio gets smaller, the base station hit ratio gets higher since there are more requests reach the base stations. Figure 19 and Figure 20 show that the query costs of all three algorithms increase dramatically with the growth of object size.

50 100 150 200 250 300 350

50 60 70 80 90 100 110 120 130 140 150 Max object size

Client query cost

CCR LRU LFU

Figure 19: Client query cost under various maximum sizes of objects

20 25 30 35 40 45 50 55 60 65

50 60 70 80 90 100 110 120 130 140 150 Max object size

Base station query cost

CCR LRU LFU

Figure 20: Base station query cost under various maximum sizes of objects

0 5 10 15 20 25

40 50 60 70 80 90

LDSProb (%)

Client hit ratio (%)

CCR LRU LFU

Figure 21: Client hit ratio under various of LDSProb 4.2.4 Impact of LDSProb

Finally we adjust the access probability of LDS to observe the impact on all three algorithms.

When LDSProb is large, the properties of LDS will be significant. So the cache replacement algorithms which take the characteristics of LDS into consideration will gain higher perfor-mance. The simulation results are shown in Figure 21, Figure 22, Figure 23, and Figure 24.

In Figure 21, we can see that the CCR performs superiorly than LRU and LFU in higher LDSProb. Since the CCR considers the properties of LDS, so when the access probability of LDS is high, the performance of CCR is ascendant. The same situation happens in Figure 22, Figure 23, and Figure 24.

5 Conclusions

In this paper, we proposed the Collaborative Cache Replacement algorithm which takes both location dependent and independent service into consideration. By the collaboration between clients and base stations, we can make the caching usage of entire environment more efficient.

We derived a profit function which considering several important factors of both location

0

Base station hit ratio (%)

CCR LRU LFU

Figure 22: Base station hit ratio under various of LDSProb

100

Figure 23: Client query cost under various of LDSProb

30

Figure 24: Base station query cost under various of LDSProb

dependent service and location independent service. By using the profit function, we can evaluate the profit of each cached service object for cache replacement. In addition to deriving profit function, we also construct a collaboration mechanism between clients and base stations.

Through this mechanism, base stations can adjust the caching priority according to the caching situation of clients in their service area. Base stations can know which data are popular and keep them in the cache. The experiment results showed that the proposed CCR is very effective and outperforms the conventional cache replacement algorithms.

References

[1] M. Abrams, C. Standridge, G. Abdulla, S. Williams, and E. Fox. Caching proxies:

Limitations and potentials. In Proceeding of the 4th International World Wide Web Conference, pages 119—133, December 1995.

[2] C. Aggarwal, J. Wolf, , and P. Yu. Caching on the World Wide Web. IEEE Transactions on Knowledge and Data Engineering, 11(1):94—107, January/February 1999.

[3] D. Barbara and T. Imielinksi. Sleepers and Workaholics: Caching Strategies for Mobile environments. In Proceeding of ACM SIGMOD, pages 1—12, May 1994.

[4] C.-Y. Chang and M.-S. Chen. Exploring Aggregate Effect with Weighted Transcoding Graphs for Efficient Cache Replacement in Transcoding Proxies. In Proceeding of the 18th International Conference on Data Engineering, February 2002.

[5] G.Cao. A Scalable Low-Latency Cache Invalidation Strategy for Mobile Environments.

IEEE Transactions on Knowledge and Data Engineering, 15(5), September/October 2003.

[6] H. Hu, J. Xu, W. Wong, D. L. B. Zheng, and W.-C. Lee. Proactive Caching for Spa-tial Queries in Mobile Environments. In Proceeding of 21th IEEE Int. Conf. on Data Engineering (ICDE ’05), April 2005.

[7] Q. Hu and D.-L. Lee. Cache Algorithms Based on Adaptive Invalidation Report for Mobile Environments. Cluster Computing, 1(1):39—48, February 1998.

[8] Q. H. J. Xu, W.-C. Lee, and D. Lee. Performance Evaluation of an Optimal Cache Replacement Policy for Wireless Data Dissemination. IEEE Transactions on Knowledge and Data Engineering, 16(1), January 2004.

[9] A. Kahol, S. Khurana, S. Gupta, and P. Srimani. A Strategy to Manage Cache Consis-tency in a Distributed Mobile Wireless Envionment. IEEE Transactions on Parallel and Distributed Systems, 12(7):686—700, July 2001.

[10] Y. B. Lin. Modeling Techniques for Large Scale PCS Networks. IEEE Comm. Magazine, 35(2):102—107, Feburary 1997.

[11] B. Liu, W.-C. Lee, and D. Lee. Distributed Caching of Multi-dimensional Data in Mobile Environments. In Proceeding of IEEE Infocom 2004, March 2004.

[12] Q. Ren and M. Dunahm. Using Semantic Caching to Manage Location Dependent Data in Mobile Computing. In Proceeding of the MOBICOM Conference, pages 210—221, 2000.

[13] S. Williams, M. Abrams, C. Standridge, G. Abdulla, and E. Fox. Removal Policy in Network Cahces for World Wide Web Documents. In Proceeding of ACM SIGCOMM, pages 293—304, 1996.

[14] S.-Y. Wu and K.-T. Wu. Dynamic Data Management for Location Based Services in Mobile Environments. In IDEAS2003: The 7th International Database Engineering and Application Symposium, pages 180—189, July 2003.

[15] J. Xu, X. Tang, and D. Lee. Performance Analysis of Location-Dependent Cache Inval-idation Schemes for Mobile Environments. IEEE Transactions on Knowledge and Data Engineering, 15(2), March/April 2003.

[16] L. Yin and G. Cao. Supporting Cooperative Caching in Ad Hoc Networks. In Proceeding of IEEE Infocom 2004, March 2004.

[17] J. Yuen, E. Chan, K. Lam, and H. Leung. Cache Invalidation Scheme for Mobile Com-puting Systems with Real-Time Data. ACM SIGMOD Record, 29(4):34—39, 2000.