B-Tree Based Trust Model for Resource Selection in Grid

Abstract

Trust management is an important issue in a grid environment where consumers and service providers are distributed geographically across autonomous administrative domains. In this paper, we propose reputation-based trust management architecture through the use of intermediaries, brokers. This architecture insists on multiple brokers in each domain. The entities (Consumers and the SPs) are distributed across these brokers, with each of these entities being associated with more than one broker. This improves the redundancy of information maintained at the broker sites, thereby improving the reliability. This also eases the network traffic at the broker sites while handling consumer requests and feedbacks. The issues rising out of such an arrangement of multiple brokers, namely the distribution of entities among the brokers and maintenance of consistency of information across the brokers, are addressed well in this paper. Trust-indices of SPs and consumers are evaluated and updated dynamically after the completion of each transaction. This enables the consumer to receive the response from the broker significantly quicker compared to other reputation-based-trust models where the trust-indices are computed at the request-time. To further improve the response time of each transaction, a B-tree indexing scheme has been proposed. Trust parameters such as satisfaction-value, number, cost and criticality of transactions, and different weights for past and recent transactions are considered for the trust evaluation. Our model shows a marked improvement in job success rate for various percentages of malicious entities. The selection query cost for each transaction is reduced thereby improving the overall response time is show-cased in this model. The impact of broker's feedback on the computation of trust-indices is also presented