Academia.eduAcademia.edu

Figure 1 – uploaded by Bin Yu

See full PDFdownloadDownload figure
Table 1. The parameters in the experiments The peers are limited to having no more than 4 neigh- bors and 16 acquaintances. Queries are sent only to and re- ferrals are given only to neighbors. After every 5 queries, each peer decides which acquaintances are promoted to be- come neighbors and which neighbors are demoted to be or- dinary acquaintances based on the latest ratings. Other pa- rameters are defined as follows,

Table 1 The parameters in the experiments The peers are limited to having no more than 4 neigh- bors and 16 acquaintances. Queries are sent only to and re- ferrals are given only to neighbors. After every 5 queries, each peer decides which acquaintances are promoted to be- come neighbors and which neighbors are demoted to be or- dinary acquaintances based on the latest ratings. Other pa- rameters are defined as follows,

Related Figures (6)

Figure 1. A trust graph generated from a query of P).
Figure 1. A trust graph generated from a query of P).
any peer. In our approach, each peer P; maintains a black- list for malicious peers who are swapped out from his ac- quaintance model. P; is in the blacklist of the correspond- ing peer at this time. Figure 3 illustrates the changes of rep- utation of P; in the whole process. The results indicate that our reputation mechanisms using either simple averaging or exponential averaging can detect the attacks from strategic peers. The reputation mechanism using exponential averag- ing is more effective in detecting these attacks, where the reputation of the malicious peer P; drops quickly from 0.89 to 0.20 during 400 cycles. The malicious peer P; may also attack the P2P search process through providing wrong re- ferrals. One solution is that every requesting peer may pe- nalize the weight for the credibility of P; so that they will have a reduced chance of contacting P; in the future. We will not discuss the details in this paper and defer this en- hancement to future work. Figure 2. Average rating distances of normal peers in the system
any peer. In our approach, each peer P; maintains a black- list for malicious peers who are swapped out from his ac- quaintance model. P; is in the blacklist of the correspond- ing peer at this time. Figure 3 illustrates the changes of rep- utation of P; in the whole process. The results indicate that our reputation mechanisms using either simple averaging or exponential averaging can detect the attacks from strategic peers. The reputation mechanism using exponential averag- ing is more effective in detecting these attacks, where the reputation of the malicious peer P; drops quickly from 0.89 to 0.20 during 400 cycles. The malicious peer P; may also attack the P2P search process through providing wrong re- ferrals. One solution is that every requesting peer may pe- nalize the weight for the credibility of P; so that they will have a reduced chance of contacting P; in the future. We will not discuss the details in this paper and defer this en- hancement to future work. Figure 2. Average rating distances of normal peers in the system
Figure 3. Reputation of an independent peer in group G
Figure 3. Reputation of an independent peer in group G
Figure 5. Average weights of witnesses for different exaggeration coefficients
Figure 5. Average weights of witnesses for different exaggeration coefficients
Figure 6. Average weights of independent or collusive peers in group G,
Figure 6. Average weights of independent or collusive peers in group G,
Related topics:
Trusted Third Party
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved