Resource Reliability in Kad

Poisoned search results are misleading people to download invalid, incorrect or even nonexistent resources in some popular file sharing systems. In order to estimate effects brought by those poisoned results, a dataset about 3 million items were gathered from Kad, including shared resources´ hash values along with related metadata. We analyzed the amount of poisoned results based on the dataset, and revealed zipf-like distributions between number of names and the number of hash values. Besides, we evaluated the indirect effects of poisoned search results through the dissimilarities among names of different hash values having same metadata. The analysis shows that, based on our dataset, audios are the most shared but least poisoned indirectly, videos are the most poisoned directly and indirectly, and other types are indirectly poisoned almost at the same level.