| Summary: | 碩士 === 國立屏東科技大學 === 資訊管理系所 === 96 === Peer-to-Peer （P2P） systems that nodes interact directly to share resources and services provide a kind platform of file sharing. Although P2P systems are convenient for getting resources, there are some problems such as inauthentic files or free-rider. In order to download good resources or solve free-riding, many researches provide “Reputation Mechanism” to quantify their behavior. It can assist us in having the references when we are going to download files and reduce the probability of accessing bad files.
A few reputation computing of the mechanisms are based on “amount of resources contribution”. The more they provided the more reputation value they have. Then they can access more resources. Such mechanisms can animate users to share files but they can not ensure the quality of resources. Others are based on “appreciation feedback”. They give some appreciations depending on the files which they downloaded. Such mechanisms can reduce the quantity of bad files. But it can not avoid the free-rider and it have to assume every appreciation is true.
In order to consider the quality of the resources, inauthentic feedback and free-riding, we propose a new mechanism to solve these questions. This mechanism includes three parts. The first part is called “The policy of dynamic accessing files”. It can not only provide the simulative effect to reduce the problem about free-rider but also provide the flexibility of accessing files for users. Users who can access resources range depend on their reputation. In order to avoid the users whose reputation values are close to give wrong appreciations, the second part of this mechanism is called “The policy of valid appreciation feedback”. This part prevents many wrong appreciations. The last part is called “Reputation management,” it can quantify the users’ behavior and provide the reference for first part and second part.
We can discover that our reputation mechanism can differentiate those people who provide good or bad resources and provide truthful or wrong appreciation. That means our mechanism can differentiate any types of users according to their behavior. Besides, we propose three measures for level dividing. The experience shows the “lower bound” efficiency is the best. Finally, we compare with other resources accessing policies. It results that our efficiency is better than them.
|
|---|
