Bittorrent Connectivity Within The Campus Network And Subnet Clusters

BitTorrent (abbr. BT) is still the most widely used P2P protocol, which accounts for a greater part of traffic in both Internet and ASes. At present, the studies about BT all over the world are deep in progress, which can be separated into several aspects, i.e. BT application at different aspects and in various fields, BT performance analysis, improvement and new algorithm presentation, modeling and simulating BT, BT characteristic identification and measurement, etc. The question how BT overlay networks look like in Internet had been answered several years ago. But in the age when various BT client apps have emerged and are emerging, whichever makes its extension to original BT protocol, in the environment moving from Internet to AS-level campus networks, the question if BT overlays change or not, retrieves its new significance.This paper is to study on BT overlay in campus networks, and its characteristic of peer nodes forming small world (clustering) on its logical overlay in the bridge-separating physical subnets of campus networks. We gathered all the traffic running inside or outside a subnet, filtered out key BT packets which were then put in order, analyzed and made discussion on. In the case that captured Ethernet frame length is limited to 400 bytes, we extracted as many interesting BT messages as possible according to protocol keywords, and computed the real numbers of the interesting messages in the original packets using a reasonable estimate algorithm.From the result of modeling and computing internal cost ratio (to evaluate the connectivity) and subnet clustering coefficient (to evaluate clustering), we conclude that physical subnets in a campus network are like subnets in a WAN, where internal cost effect exists, but BitTorrent doesn't make subnet cluster apparently. Additionally, from the statistics of file sub-piece downloading, we can see that in the range of a file, the numbers of peer uploading pieces correspond to a non-uniformity and non-linear distribution, i.e. they are not selected simply at random.