iTorrent Model
Partially Centralized
Apple servers will be responsible for setting up trackers, these trackers will only be available through the iTunes store to users who buy the movies. The Apple server will set up users with other users who own or are downloading the same content.
The client will be integrated into iTunes, and will be transparent to the user. The idea is that the user will not know whether they are downloading from the iTunes server or from their peers (the give-away will be upload rates).
Changes to the BitTorrent Model
DRM and licensing – users will only share with other users who already have licenses for the content they are downloading. Still, the files will be encoded with Apple DRM to enforce the iTunes licensing agreements.
Peers will be able to opt-out of sharing, and control how much they share. Instead of forcing users to share, there will be an incentive system. This means all users will share blindly (in regard to peer cooperation) and not employ tit-for tat.
Reporting system – the iTorrent client will be responsible for tracking and reporting how much a user has shared, which will translate into rewards.
iTunes Server Supplementation
After research and experimentation, it has become increasingly apparent that, under the status quo of disk space and bandwidth, the BitTorrent model will only supplement and not replace the iTunes central server model, at least for now. Due to disk space constraints (most users have only about 100GB of hard drive space, and with movies ranging from 1-2 GB and HD movies upwards of 19 GB, users cannot currently afford to keep these files in their iTunes libraries, and will therefore move them to auxiliary storage devices (DVDs, external hard drives), therefore removing nodes from the network and reducing the density of file availability. At this point in time, bandwidth, except in local, high-bandwidth networks (such as college campuses), is too low to support the high speed of downloads expected of iTunes without a huge number of contributors. These issues will ameliorate in the future, as disk space continues to become cheaper and ISPs increase bandwidth.
For now, iTunes will have to support the iTorrent system in order to maintain the high quality of service users have come to expect. In my experiments, iTunes movies download at an average rate of 1.69 mb/s, while BitTorrents download at only 95 kb/s. While a 1.4GB file downloaded from the iTunes server took only 15 minutes to download, a 700MB file from BitTorrent (with upwards of 450 sources) took over 2 hours.
If we assume that every user can upload at about 28 kb/s (this is a fairly modest amount for most broadband connections), we find we need 60 users sharing at this rate to allow a 1,000 megabyte file to be downloaded in 10 minutes (approximately the rate of the iTunes server). If we increase the target download time to 15 minutes for the same file, users must only share at 19 kb/s. This speed (1.1mb/s) is also reasonable.
The client will therefore monitor its connection speed, and when it falls to an unacceptable transfer rate, will contact the server and the server will enter the torrent as another peer to supplement the download process. So when there are 60 users uploading at 28kb/s, the server will not be doing any work (other than tracker monitoring). When the number of users drops to 50 or the average upload rate drops, the server will supplement the download process.
What this amounts to is that, the more people that are downloading a file, the more iTorrent helps the server. The upshot is that iTorrent will reduce server burdens when the burden on the server would be at its highest. So when new films come out or are in high demand for any reason, the iTorrent system takes on a larger amount of the file transfers.
The result is that the iTorrent system will not replace the central server, but it will supplement it at the times it is needed the most, to reduce overall burden and keep server and bandwidth costs lower for iTunes.
Experiment Results
iTunes average download rate: 1.69 mb/s
BitTorrent average transfer rate: 95 kb/s
For detailed experiment results, download the detailed results (excel).
This document also includes calculations for target download rates.
Resources
[1] Bram Cohen. Incentives Build Robustness in BitTorrent. http://www.bittorrent.org/bittorrentecon.pdf
- discusses peaks and exponential drop-off of nodes
[2] M. Izal, G. Urvoy-Keller, E.W. Biersack, P.A. Felber, A. Al Hamra, and L.Garces-Erice. Dissecting BitTorrent: Five Months in a Torrent's Lifetime.
- discusses client participation
[3] Lei Guo, Songqing Chen, Zhen Xiao, Enhua Tan, Xiaoning Ding, and Xiaodong Zhang. A Performance Study of BitTorrent-like Peer-to-Peer Systems
- existing BitTorrent system provides poor service availability, fluctuating downloading performance due to exponentially decreasing peer arrival rate. Suggests inter-torrent collaboration
[4] Modeling and performance analysis of BitTorrent-like peer-to-peer networks http://delivery.acm.org/10.1145/1020000/1015508/p367-qiu.pdf?key1=1015508&key2=5906082711&coll=GUIDE&dl=GUIDE&CFID=12510790&CFTOKEN=96054971
- derived mathematical expressions for average seeds, downloaders, download time, etc, as functions of parameters such as peer arrival rate, downloader leaving rate, seed leaving rate, uploader bandwidth
[5] http://torrentfreak.com/optimize-your-bittorrent-download-speed/
- Discussion of how to optimize BitTorrent speeds
[6] Enhancing BitTorrent with Trust Management. http://disl.cc.gatech.edu/courses/cs8803h_spring2005/project/proposals/lenfest.pdf
[7] Analyzing and Improving BitTorrent Performance http://www.sics.se/~sameh/research/P2P/BiTorrent/Analyzing%20and%20Improving%20BitTorrent%20Performance/msr-tr-2005-03.pdf
http://www.cs.cmu.edu/~ashu/papers/msr-tr-2005-03.pdf
[8] A Measurement Study of the BitTorrent Peer-to-Peer File-Sharing System http://www.pds.ewi.tudelft.nl/~pouwelse/bittorrent_measurements.pdf
- discusses the drop-off of available seeds after given time periods
[9] Towards Digital Rights Protection in BitTorrent-like P2P Systems http://www.list.gmu.edu/zhang/pub/bt.pdf
- discusses implementation of DRM in BitTorrent model
[10] How to Solve Copyright Problems in P2P Content Distribution? http://www.valimaki.com/org/mostbook.pdf
- Discussion on DRM distribution systems.
[11] Data Integrity and Proof of Service in BitTorrent-like P2P Environments http://www.cs.uoregon.edu/~lijun/pubs/junli_IATAC06_sub.pdf
- Discusses issues of users not sharing
Partially Centralized
Apple servers will be responsible for setting up trackers, these trackers will only be available through the iTunes store to users who buy the movies. The Apple server will set up users with other users who own or are downloading the same content.
The client will be integrated into iTunes, and will be transparent to the user. The idea is that the user will not know whether they are downloading from the iTunes server or from their peers (the give-away will be upload rates).
Changes to the BitTorrent Model
DRM and licensing – users will only share with other users who already have licenses for the content they are downloading. Still, the files will be encoded with Apple DRM to enforce the iTunes licensing agreements.
Peers will be able to opt-out of sharing, and control how much they share. Instead of forcing users to share, there will be an incentive system. This means all users will share blindly (in regard to peer cooperation) and not employ tit-for tat.
Reporting system – the iTorrent client will be responsible for tracking and reporting how much a user has shared, which will translate into rewards.
iTunes Server Supplementation
After research and experimentation, it has become increasingly apparent that, under the status quo of disk space and bandwidth, the BitTorrent model will only supplement and not replace the iTunes central server model, at least for now. Due to disk space constraints (most users have only about 100GB of hard drive space, and with movies ranging from 1-2 GB and HD movies upwards of 19 GB, users cannot currently afford to keep these files in their iTunes libraries, and will therefore move them to auxiliary storage devices (DVDs, external hard drives), therefore removing nodes from the network and reducing the density of file availability. At this point in time, bandwidth, except in local, high-bandwidth networks (such as college campuses), is too low to support the high speed of downloads expected of iTunes without a huge number of contributors. These issues will ameliorate in the future, as disk space continues to become cheaper and ISPs increase bandwidth.
For now, iTunes will have to support the iTorrent system in order to maintain the high quality of service users have come to expect. In my experiments, iTunes movies download at an average rate of 1.69 mb/s, while BitTorrents download at only 95 kb/s. While a 1.4GB file downloaded from the iTunes server took only 15 minutes to download, a 700MB file from BitTorrent (with upwards of 450 sources) took over 2 hours.
If we assume that every user can upload at about 28 kb/s (this is a fairly modest amount for most broadband connections), we find we need 60 users sharing at this rate to allow a 1,000 megabyte file to be downloaded in 10 minutes (approximately the rate of the iTunes server). If we increase the target download time to 15 minutes for the same file, users must only share at 19 kb/s. This speed (1.1mb/s) is also reasonable.
The client will therefore monitor its connection speed, and when it falls to an unacceptable transfer rate, will contact the server and the server will enter the torrent as another peer to supplement the download process. So when there are 60 users uploading at 28kb/s, the server will not be doing any work (other than tracker monitoring). When the number of users drops to 50 or the average upload rate drops, the server will supplement the download process.
What this amounts to is that, the more people that are downloading a file, the more iTorrent helps the server. The upshot is that iTorrent will reduce server burdens when the burden on the server would be at its highest. So when new films come out or are in high demand for any reason, the iTorrent system takes on a larger amount of the file transfers.
The result is that the iTorrent system will not replace the central server, but it will supplement it at the times it is needed the most, to reduce overall burden and keep server and bandwidth costs lower for iTunes.
Experiment Results
iTunes average download rate: 1.69 mb/s
BitTorrent average transfer rate: 95 kb/s
For detailed experiment results, download the detailed results (excel).
This document also includes calculations for target download rates.
Resources
[1] Bram Cohen. Incentives Build Robustness in BitTorrent. http://www.bittorrent.org/bittorrentecon.pdf
- discusses peaks and exponential drop-off of nodes
[2] M. Izal, G. Urvoy-Keller, E.W. Biersack, P.A. Felber, A. Al Hamra, and L.Garces-Erice. Dissecting BitTorrent: Five Months in a Torrent's Lifetime.
- discusses client participation
[3] Lei Guo, Songqing Chen, Zhen Xiao, Enhua Tan, Xiaoning Ding, and Xiaodong Zhang. A Performance Study of BitTorrent-like Peer-to-Peer Systems
- existing BitTorrent system provides poor service availability, fluctuating downloading performance due to exponentially decreasing peer arrival rate. Suggests inter-torrent collaboration
[4] Modeling and performance analysis of BitTorrent-like peer-to-peer networks http://delivery.acm.org/10.1145/1020000/1015508/p367-qiu.pdf?key1=1015508&key2=5906082711&coll=GUIDE&dl=GUIDE&CFID=12510790&CFTOKEN=96054971
- derived mathematical expressions for average seeds, downloaders, download time, etc, as functions of parameters such as peer arrival rate, downloader leaving rate, seed leaving rate, uploader bandwidth
[5] http://torrentfreak.com/optimize-your-bittorrent-download-speed/
- Discussion of how to optimize BitTorrent speeds
[6] Enhancing BitTorrent with Trust Management. http://disl.cc.gatech.edu/courses/cs8803h_spring2005/project/proposals/lenfest.pdf
[7] Analyzing and Improving BitTorrent Performance http://www.sics.se/~sameh/research/P2P/BiTorrent/Analyzing%20and%20Improving%20BitTorrent%20Performance/msr-tr-2005-03.pdf
http://www.cs.cmu.edu/~ashu/papers/msr-tr-2005-03.pdf
[8] A Measurement Study of the BitTorrent Peer-to-Peer File-Sharing System http://www.pds.ewi.tudelft.nl/~pouwelse/bittorrent_measurements.pdf
- discusses the drop-off of available seeds after given time periods
[9] Towards Digital Rights Protection in BitTorrent-like P2P Systems http://www.list.gmu.edu/zhang/pub/bt.pdf
- discusses implementation of DRM in BitTorrent model
[10] How to Solve Copyright Problems in P2P Content Distribution? http://www.valimaki.com/org/mostbook.pdf
- Discussion on DRM distribution systems.
[11] Data Integrity and Proof of Service in BitTorrent-like P2P Environments http://www.cs.uoregon.edu/~lijun/pubs/junli_IATAC06_sub.pdf
- Discusses issues of users not sharing
No comments:
Post a Comment