Analysis of Internet Music Content Distribution

1
Analysis of Internet Music Content Distribution
DIMI Project - UCLA/Warner Bros.
Wendy Aylsworth Charles L. Dages Warner Bros.
Vince Busam Sasha Slijepcevic Miodrag
Potkonjak Richard Muntz UCLA
2
Full Project Objectives

• Statistical analysis of web and user behavior
  toward pirated audio material.
• Evaluation of watermarking and fingerprinting
  techniques.
• Determine methods to control distribution of
  digital content.
• Processes and procedures
• Technical controls
• Legal announcements and controls

3
Immediate Project Tasks

• Develop experiment for sampling web data and user
  behavior
• Data acquisition
• Data analysis

4
Goals

• To supply real numbers to the discussion of the
  content distribution on the Internet
• To identify interesting data about the Internet
  users who exchange MP3 files
• To develop tools for acquiring interesting
  information
• Exploit existing services
• DNS (mapping URLs into IP addresses and vice
  versa)
• whois (returns the owner of an IP address)

5
Outline

• Current target data (comments welcomed)
• Tools
• MP3 on FTP
• Napster
• Future work

6
Target Data

• Who are MP3 users?
• Are they mostly college students?
• How many are outside of US?
• What types of connection they have DSL, cable
  modem, Ethernet from dorms?
• How many songs they offer for download?
• Who are the most popular artists?

7
Target Data

• How soon do files show up after or even before
  being officially released?
• Who are the users who are first to convert CDs
  into MP3 files?
• If pirated copies appear before official release,
  do they originate at the same subset of IP
  addresses?
• How many different MP3 versions of a same song
  exist?
• Different versions indicate different pirate
  sources

8
Internet Access in USby Type of Connection
Internet TV 2.2
DSL 0.3
Cable 4.5
Dial-Up Access 93
From TRIs report, May 2000
9
Getting Information About Users
137.132.94.4
10
Getting Information About Users
137.132.94.4

• We want to find out using DNS and whois
• Owner of the IP address (college, ISP)
• Connection type (DSL, Cable modem, Dial-up)
• Geographic location (Europe, )

11
MP3 on FTP

• FTP is the traditional method of MP3 sharing
  (before Napster)
• Search engines crawl FTP servers, indexing
  available MP3 files
• mp3.lycos.com
• 2Look4.com
• oth.net
• Users go to a search engine, look for an
  artist/song, then connect to the FTP sites to
  download

12
FTP Crawler

• Gather a list of FTP sites by entering search
  terms into popular FTP search engines

http//mp3.lycos.com
List of FTPservers
Search terms
http//2Look4.com
http//oth.net

• Crawl each of these servers, gathering a list of
  all files offered for download
• Re-crawl all servers weekly, to allow future
  analysis

13
Dynamic IP Addresses

• FTP servers whose IP address changes (dial-up
  access) use the dynamic DNS
• mp3.dhs.org/mp3/Artist-SongTitle.mp3 is a typical
  file
• mp3.dhs.org does not give any of target data
• Using DNS, turn the dynamic domain name into an
  IP address

mp3.dhs.org
137.132.94.4
14
DNS lookup sequence

• Using reverse DNS, turn IP addresses into native
  hostnames if possible

sun450.comp.nus.edu.sg
137.132.94.4

• This shows that the FTP server is in Singapore
• More specifically, at School of Computing at
  National University of Singapore
• Not all IP addresses have corresponding DNS name

15
Whois lookups

• If reverse DNS does not submit any answer, we try
  to find the owner of an IP address using whois
  lookup at ARIN
• American Registry for Internet Numbers
• Authority for IP addresses in US
• Will return owner of each IP address
• whois 209.185.207.136_at_whois.arin.net
• Returns Flashcom owned netblock
• DSL connection

16
Locations of FTP Servers

• 3800 servers in April, 390 servers in July
• Reasons Napster, firewalls at cable modem
  networks, summer break

7
13
28
18
43
41
2
22
22
5
17
Files by Connection Type

• 230,000 MP3 files in April, 340,000 in July,
  mostly illegal by visual inspection of filenames
• 58 sites with over 1000 MP3s in April, 119 in
  July
• 8500 MP3s on a single site in Canada

2
8
26
26
40
52
12
6
26
2
APRIL
JULY
18
Napster

• The most popular application for sharing MP3s
• Developed by San Mateo based Napster, Inc.
• Has been imitated by many competitors, but still
  has largest market share

19
Napster

• Client-client transfer
• Clients register with a server
• Clients then transfer files between each other
• Previously done with Email, ICQ, IRC
• Napster offers easy package, allowing this to
  surpass client-server transfers in number of users

20
Napster Architecture

• CONNECT
• IP address
• Shared files

21
Napster Architecture

• REQUEST
• Search term(s)

22
Napster Architecture

• RESPONSE
• IP addresses of users with files containing
  search terms
• File names

23
Napster Architecture
Actual File Transfer
137.132.94.4
24
Napster Tools

• Napster protocol has been reverse-engineered
• Modified GTK-Napster (Open Source clone) to log
  IP addresses of servers returned in searches
• Napster's application forces users to share their
  download directory
• Most users will be registered as servers
• Users can move MP3s out of a shared directory
• Use unapproved clone without sharing
• Application continues to run even if the main
  window is closed - oblivious to many users

25
Napster

• Our data is a sample of connected users
• Found 63,000 files on 24,000 unique servers

4
16
18
51
9
2
26
Pending Work

• Gather more information from Napster
• Snapshots of searches by time
• File format returned by search (bitrate, size...)
• Bandwidth of the connected clients measured
  sending ping packets while downloading
• Correlate our data with other studies
• Compare MP3 transfer to record sales near
  colleges in last 2 years
• Study by VNU Entertainment Marketing Solutions
  (4 decrease in stores near colleges)
• Watch MP3 usage as SDMI becomes popular

27
Pending Work (contd)

• Inject files with false metadata (title, artist,
  ID tag)
• Napster
• Set up Web or FTP server
• Measure how far a false version of a song spreads
  on the Internet
• Develop techniques for fast recognition of a song
• How to prove that the song is the one that the
  title claims without downloading the whole file