frequencies, connection probabilities and cache size when performing hoarding ... bile clients to hoard every data object available on the net- work. This leads to ...
Improving Data Accessibility For Mobile Clients Through Cooperative Hoarding Kwong Yuen Lai Zahir Tari Peter Bertok School of Computer Science and Information Technology RMIT University, Melbourne, Australia Email: kwonlai, zahirt, pbertok@cs.rmit.edu.au Abstract In this paper, we introduce the concept of cooperative hoarding to reduce the risks of cache misses for mobile clients. Cooperative hoarding takes advantage of group mobility behaviour, combined with peer cooperation in adhoc mode, to improve hoard performance. Two cooperative hoarding approaches that take into account clients’ access frequencies, connection probabilities and cache size when performing hoarding are proposed. Test results show that the proposed methods significantly improve cache hit ratio and reduce query costs compared to existing approaches.
Greedy Global Hoarding (GGH) The GGH method improves the effectiveness of hoarding by providing each client with the knowledge of what other clients within the group have already hoarded. The algorithm is outlined below:
1 Introduction Due to limited storage capacity, it is impossible for mobile clients to hoard every data object available on the network. This leads to the problem of critical cache misses during disconnected operations. The penalty of such cache misses is very high and may prevent a disconnected client from continuing its operation altogether. To reduce the likelihood of cache misses for mobile clients, the concept of cooperative hoarding is proposed in this work. Recent research have shown that mobile users often move in groups [4]. Cooperative hoarding takes advantage of the fact that even when disconnected from the network, clients may still be able to communicate with each other in ad-hoc mode. By performing hoarding cooperatively, clients can share their hoard content during disconnections to achieve higher data accessibility and reduce the risk of critical cache misses.
2 Proposed methods We consider a mobility group that consists of mobile clients. A set of data objects (denoted ) is available for client access from a central server. It is assumed the clients are currently in an area with strong connectivity and have access to the server, but are planning to travel together to an area with limited or no network coverage.
Proceedings of the 21st International Conference on Data Engineering (ICDE 2005) 1084-4627/05 $20.00 © 2005 IEEE
1. Calculate the global connection probability of each client. is defined as :
(1)
where is the connection probability between two clients and . 2. The client with the highest connection probability (say ) is chosen as the starting node. 3.
calculates a cost value for each data object in . The value of represents the average penalty paid per query if object is not cached locally by , and is calculated as follows :
(2)
where is the probability of querying object and is the cost of a cache miss.
4. Given for all in , fills its cache with objects starting from the one with the highest until its cache is full. 5.
then constructs a matrix, to indicate which objects it has if is cached by and = hoarded.
0 otherwise. is sent to the next client in a breadth first traversal through . For each client traversed, the following steps are repeated :
6. When a client (say ) receives , it recalculates for all objects with the help of as follows: cost
min
if where
otherwise
where is the cost for from another client .
to fetch an object
7. The objects with the highest are cached by until its cache is full.
No Cooperation With Cooperation DAFN GGH CAP
80
Cache hit ratio
8. Once has finished filling its cache, it updates and passes it to the next client in the breadth first traversal order. The process is repeated from step 6 for the newly selected client. This continues until every client in the group has filled their caches.
100
60
40
Cooperative Access Probability-based hoarding (CAP) The second proposed method CAP finds the best location to cache each object so the global access cost is minimised.
20
1. The average access probability of each object in is defined as follows:
0
2. Let denote the global access cost if object is cached by a client :
(4)
where is the access cost for client to access object from a peer . 3. For each object , starting from the one with the highest value, a client is identified such that min for all in . 4. represents the best location to store , which minimises the total cost to access for the whole group. A copy of is placed in the cache of . 5. Steps 3 and 4 are repeated until the caches of all the clients in the group are filled.
3 Simulation Results The average cache hit ratio of GGH and CAP is shown against the number of clients in Figure 1. As the number of clients increases, all the simulated approaches, except for ‘No Cooperation’ achieve increasing cache hit ratio. The reason for this is that as the number of clients increases, clients are likely to be connected to more peers, which gives them better chances of finding the object they need from a peer. It is found that even when there are only a few clients, cooperation significantly improve performance.
4 Conclusion In this work, the concept of cooperative hoarding is introduced to improve data accessibility for mobile clients.
5
10
15
20
25
30
Number of clients
Figure 1. Cache Hit Ratio vs. Number of clients
(3)
where is the probability of accessing .
0
Two cooperative hoarding schemes, GGH and CAP, have been proposed. GGH improves hoard performance by allowing clients to take advantage of what their peers have hoarded when making their own hoarding decisions. On the other hand, CAP selects the best client in the group to hoard each object to maximise the number of unique objects hoarded and minimise access cost. Simulation results show that compare to existing schemes, GGH and CAP improves cache hit ratio by between 9% to 22% and are effective in supporting mobile clients during disconnections.
Acknowledgement This project is supported by the ARC (Australian Research Council - under the Linkage-Project scheme, no. LP0455234) and SUN Microsystems grant no.7832030217-AUS.
References [1] T. Hara. Replica allocation methods in ad hoc networks with data update. Mobile Networks and Applications, 8:343–354, 2003. [2] T. Hara, N. Murakami, and S. Nishio. Replica allocation for coreelated data items in ad hoc sensor networks. SIGMOD Record, 33(1):38–43, March 2004. [3] J. Huang, M. Chen, and W. Peng. Exploring group mobility for replica data allocation in a mobile environment. In Proceedings of the ACM Conference on Information and Knowledge Management, pages 161–168, November 2003. [4] K. H. Wang and B. Li. Group mobility and partition prediction in wireless ad-hoc networks. In Proceedings of the IEEE International Conference on Communication (ICC’02), pages 1017–1021, April 2002. [5] L. Yin and G. Cao. Supporting cooperative caching in ad hoc networks. In Proceedings of the IEEE INFOCOM, March 2004.
Proceedings of the 21st International Conference on Data Engineering (ICDE 2005) 1084-4627/05 $20.00 © 2005 IEEE
