Characterization of shared files in peer-to-peer network

Abstract

Today most peer-to-peer file sharing systems rife with polluted files. Actual content of polluted file is different from what user expected based on file metadata such as name or size. File pollution decreases the utility of peer-to-peer shared networks by disturbing peers to locate desired files. In this thesis, we analyze characteristics of shared files in eDonkey P2P networks in terms of lifetime, incompleteness, and pollution: how long files are available for download, whether peers download any file completely, and what kind of pollution occurs. For our analysis, we use {\\it eDonkey history}, one month archive of hourly snapshots on file states, for 328,515 files. It was recorded for five months from November 9, 2004 to April 13, 2005. File lifetime analysis shows that files are available for relatively long period of time; lifetime of 45% of files is longer than two months. We observe long-incomplete download problem that files cannot be downloaded completely over one month even though there are available sources. Small number of long-incomplete files, 1.47% of total files, generates about 17% of total requests to find uploading sources. About 32% of polluted files is mislabeled of adult contents and 9.3% of them is virus-infected. Polluted files are available longer than others; average lifetime of polluted files is about 11 weeks.