I have been working, time permitting, on my BitTorrent implementation over the past couple of months. I’d like to note first of all that a fellow
BitTorrent hacker, the maintainer of LibBt, Ryan Walklin, contacted me about some of the issues I was writing posts about. Unfortunately I don’t think I was of much help to him with his MSE/OpenSSL problems, but it was a very useful contact for me since Ryan took the time to explain some of the downloading algorithms used in LibBT.
It was great to get some answers to some of my questions, thanks for that Ryan! The discussion made me all the more aware of the limitations in my design, and I have since been re-structuring much of the network code. This correspondence alone proves to me that its worth taking the time to write posts on some of the issues I’m working through. Today I finally got where I want to be in terms of keeping track of in-progress block transfers and peer connections and so forth, so I feel I’m making good progress. However for a long time now I’ve been wishing for some sort of analysis tool to see exactly what messages existing clients (which work!) are sending, and in what order. Just now I discovered an “instrumented” version of the mainline BitTorrent client, from the folks at INRIA. I’d been aware of their work in BitTorrent for some time, having read their papers - which are responsible for me using the rarest-first piece download strategy. Anyway, this patched client they have published is truly eye-opening. It shows exactly the messages the client is sending and receiving, along with timing and internal state information. Absolutely invaluable for seeing how a real world transfer should go! This information should greatly help me in making my implementation much more robust.
Tags: BitTorrent, C, UNIX
Related posts: