Trust in context of resource sharing, e.g. Grid.

1.	Description:

Suppose in a Grid environment, users share resources for others’ usage. Each user can contribute to the Grid by sharing resource. At the same time, users are free to utilize others’ resources to solve certain problems. Hence, each user is treated equally as a peer, though in one instance of resource usage, a peer takes the role of resource provider and another takes the role of consumer. 

Currently, Grid infrastructure already provides mechanisms to describe resources in order to facilitate the resource selection. However, existing mechanisms generally do not provide anything about the resource providers’ competence and intention. It is possible that a particular provider, after agreeing to provide resource to solve the consumer’s problem, fails to fulfill the agreement. A provider’s failure to fulfill the agreement may be caused by many reasons, e.g. some other potential consumers offer higher price for resource usage, if the resource usage is with certain monetary compensation. The real reason of failure is beyond the scope of trust modeling, whereas we care more about using trust to model users’ behavior (in providing resources) in order to facilitate a more robust resource selection. Here, “robust” means that the resource selection does select a provider that will not fail to solve the problems. 

2.	Trust metrics

After each resource usage, the resource consumer gives a rating for the provider as a feedback of the provider’s behavior. The feedback can be binary rating or probabilistic rating. 

Then before EACH new resource usage, the consumer evaluates each candidate provider’s trustworthiness based on the latter’s previous feedbacks. The derived trustworthiness serves as an indicator whether the corresponding provider will fail in new instances of resource usage.

3.	Evaluation of Trust metrics

We can assume that whether a peer will fail when acting as a provider in one instance of resource usage is controlled by its **loyalty** to the Grid. Peer’s loyalty is generally hidden from others. Trustworthiness, as an indicator whether the peer will fail in new instances of resource usage, is basically a prediction of the corresponding peer’s loyalty. One evaluation of trust metrics is that the derived trustworthiness must be close to loyalty.

Yes, it is good that the derived trustworthiness can predict the loyalty accurately. But is it really necessary? Peers measure others’ trustworthiness for making the decision of resource selection. The derived trustworthiness may not be accurate enough. But it should be able to help peers select the competent providers and avoid the less competent ones. 

*****So is it reasonable to evaluate the trust metrics in terms of failed resource usage? ***** For example, assume that the peer keeps selecting resources to solve its problem before it is solved successfully, then the trust metrics can evaluated and compared by the numbers of re-selection.