Digging into KAD users' shared folders

Characterizing peer-to-peer overlays is crucial for under- standing their impact on service provider networks and as- sessing their performance. Most popular le exchange appli- cations use distributed hash tables (DHTs) as a framework for managing information. Their fully decentralized nature makes monitoring and users tracking challenging. In this work, we analyze KAD, a widely deployed DHT system. Thanks to the unique possibility to monitor a large popu- lation of about 20,000 ADSL clients at the edge of the net- work, we are able to characterize the content downloaded and shared by local users. We devised a passive content monitoring toolkit to reliably track users between sessions despite dynamic IP allocation. We applied our tool over one month of data. Our main ndings are: (i) Over half a TB of fresh data is downloaded every day by the users we monitor, (ii) A signicant fraction of peers (20%) regulary change their ID in the KAD overlay, either on a session ba- sis or on a sub-session basis, which can be detrimental to the proper functioning of the DHT, (iii) Those users, that we term Chameleon users, are connected longer than regu- lar users, and they (claim to) have less data in their shared folder than regular peers and (iv) As a consequence, even a non biased observation of the users shared folder can only provide a lower bound of the content downloaded and shared by a population of ADSL users.