Computer Science
cs670
The University Of Alabama in Huntsville
BitTorrent
G. W. Cox – Spring 2005
What it is
Computer Science
cs670
• A p-t-p file sharing system with some novel
features:
The University Of Alabama in Huntsville
– Load sharing through file splitting
– Uses bandwidth of peers instead of a server
(distributes bandwidth costs)
– A concept of “fairness” among peers
• Successfully used:
– Used to distribute RedHat 9 ISOs (about 80TB)
– Lindows: 50% price break for downloading OS via
BitTorrent
G. W. Cox – Spring 2005
1
The idea (setup)
Computer Science
cs670
• A “seed” node has the file
The University Of Alabama in Huntsville
• File is split into fixed-size segments (256KB
typ)
• Hash calculated for each segment
• A “tracker” node is associated with the file
• A “.torrent” meta-file is built for the file –
identifies the address of the tracker node
• The .torrent file is passed around the web
G. W. Cox – Spring 2005
The idea (download)
Computer Science
cs670
• A client wants to download the file
• Contacts the tracker identified in the .torrent file (using HTTP)
The University Of Alabama in Huntsville
• Tracker sends client a (random) list of peers who have/are
downloading the file
• Client contacts peers on list to see which segments of the file
they have
• Client requests segments from peers (via TCP)
• Client uses hash from .torrent to confirm that segment is
legitimate
• Client reports to other peers on the list that it has the segment
• Other peers start to contact client to get the segment (while
client is getting other segments)
G. W. Cox – Spring 2005
2
The flow
Computer Science
cs670
Segments
Original 5
The University Of Alabama in Huntsville
4 Seed
File 3 Node
2
1
3 Tracker
1 Node
Client
Node
3
New
Client 1 Client
Node
G. W. Cox – Spring 2005
Load sharing
Computer Science
cs670
• As the file segments are downloaded by
The University Of Alabama in Huntsville
more and more peers, the peers
become the sources for further
downloads
• Because the tracker randomizes the list
of peers, the load gets spread randomly
G. W. Cox – Spring 2005
3
Fairness
Computer Science
cs670
• To spread the load, clients have to have an incentive
to allow peers to download from them – BitTorrent
The University Of Alabama in Huntsville
does this with an economic system
• Each peer A keeps a record of the peers it has gotten
segments from
• When a peer B asks to download from A:
– A checks its list
– If B has given A segments in the past, B’s request gets a
higher priority
– Else, B’s request gets a low priority
• The net result is a “tit-for-tat” arrangement – in order
to be a successful downloader, a peer must allow
frequent uploading
G. W. Cox – Spring 2005
Additional load spreading
Computer Science
cs670
• When a client needs to choose which segment to
request first, a common approach is “rarest-first”
The University Of Alabama in Huntsville
• In rarest-first approach, client identifies the segment
held by the fewest of its peers and tries to download
that segment
• This tends to keep the number of sources for each
segment as high as possible, speading load
• Rarest-first also gives the client an “attractive”
segment, making it a more popular source, and
giving it more “credits” to download more segments
G. W. Cox – Spring 2005
