1. Peer-to-Peer Information Systems Week 2: File Sharing
Old Dominion University
Department of Computer Science
CS 495/595 Fall 2003
Michael L. Nelson
9/2/03

2. File Sharing Milieu
we make a slightly artificial distinction between file sharing and file storage
sharing has less temporal guarantees than storage
file sharing projects are covered in a separate lecture
Popularity
KaZaA
Napster
Gnutella
Freenet
Free Haven
OceanStore
Publius
Privacy
Intermemory
Persistence

3. Where is Napster in Our Book?
Who will write for Napster?
Shawn Fanning?
incidentally, a Bacon # of 2:
Kasaras, “Music in the Age of Free Distribution: MP3 and Society”
should have been a chapter in our book

4. What is an MP3? Motion Picture Experts Group 3 layers of audio
from:
Layer 1 : 192 Kbps per channel, simple implementation, no temporal masking.
Layer 2 : 128 Kbps per channel, medium complexity, some temporal masking.
Layer 3 : 64 Kbps per channel, complex implementation, temporal masking.
roughly 1/12 the size of WAV files
1 cd: ~ 700 MB or 80 minutes
8.75 MB / minute
mp3: ~ 960 minutes in the same 700 MB!
0.72 MB / minute
instead of 16 five minute songs on a CD, you know have 192!
an order of magnitude improvement

5. Disruptive Technologies
Computers
steady (or falling) price points
increasing computational capabilities
increasing storage capabilities
Networking
high speed at home, school, work
File
an order of magnitude improvement

6. Innovator’s Dilemma Clayton M. Christensen
order of magnitude changes are disruptive technologies
disruptive technologies change the community

7. OpenNap http://opennap.sourceforge.net/ Protocol
format of client<->server communication:
<length><type><data>
100+ types!!!
client<->client communication
upload, download, firewalled upload, firewalled download
firewalled clients use the server as a proxy
session walkthrough at:

8. Gnutella as described in chapter 8
super nodes to be discussed in “Performance” lecture
really a protocol that is supported by many applications
servent == client and server

9. Gnutella Protocol Descriptor Description Ping
used to discover servents on the network
Pong
Response to a ping; describes the answering servent
Query
keyword search of a servent’s holdings
QueryHit
Response to Query; provides information about how to retrieve a file
Push
Allows firewalled servents to share files
Get
Get a file from a servent
adapted from:

10. Client-Server Cocktail Party (pp. 98-99)
Napster
Enter the party of 35M via the foyer
Connect, upload your list of files, your list is indexed at the central server
Your only friend is the party host
Napster tells you your list of files was recvd
You ask the host where the snacks are
You send your keyword search to Napster’s central server
The host points you to where the snacks are
Napster server tells you where the files are that match your query
You go to the snacks
You select a file and the Napster server brokers a download from the machine that holds it

11. Gnutella Cocktail Party (p. 98)
Enter the party, say hello to the first person you meet
Connect to any Gnutella host and issue a PING
Shortly, your friends see you and say hello
Your PING is broadcast to hosts “nearby”, who eventually respond with a PONG
You ask your friends where the snacks are
You query the hosts you “know” (via PONGs) for a file
Nobody knows, so they ask people next to them, and so on until everyone in the room has been asked.
If a host has the file, it responds to you. But the query is passed on other hosts as well until the whole network is canvassed.
The few folks near the snacks let you know where they are.
Several hits are routed back to you.
You walk over to the snacks.
You select from the hits that are presented to you.

12. Gnutella Messages
Every message has a 128-bit universal unique identifier (UUID)
Leach & Salz, “UUIDs and GUIDs”
hash of (among others) clock + MAC address
more on this in the “Naming” lecture
default time-to-live (TTL) of 7

13. Gnutella Messages
when a mesg is recvd, (temporarily) store its UUID, then forward to peers
do not forward messages with UUIDs that you have seen before
every servent decrements the TTL each time they pass the message on
when TTL==0, don’t forward the message
Pongs & QueryHits are routed directly back to the originator and not forwarded

14. Gnutella search mechanism
Steps:
Node 2 initiates search for file A 7 1 A 4 2 6 3 5
Slide from Matei Ripenau;

15. Gnutella search mechanism
Steps:
Node 2 initiates search for file A
Sends message to all neighbors 7 1 4 2 A 6 3 A 5
Slide from Matei Ripenau;

16. Gnutella search mechanism
Steps:
Node 2 initiates search for file A
Sends message to all neighbors
Neighbors forward message 7 1 4 2 A 6 3 A 5
Slide from Matei Ripenau;

17. Gnutella search mechanism
Steps:
Node 2 initiates search for file A
Sends message to all neighbors
Neighbors forward message
Nodes that have file A initiate a reply message 7 1 4 2 A 6 3
A:5 5 A
Slide from Matei Ripenau;

18. Gnutella search mechanism
Steps:
Node 2 initiates search for file A
Sends message to all neighbors
Neighbors forward message
Nodes that have file A initiate a reply message
Query reply message is back-propagated 7 1 4 2
A:7
A:5 6 3 A 5 A
Slide from Matei Ripenau;

19. Gnutella search mechanism
Steps:
Node 2 initiates search for file A
Sends message to all neighbors
Neighbors forward message
Nodes that have file A initiate a reply message
Query reply message is back-propagated 7 1 4
A:7 2
A:5 6 3 5
Slide from Matei Ripenau;

20. Gnutella search mechanism
Steps:
Node 2 initiates search for file A
Sends message to all neighbors
Neighbors forward message
Nodes that have file A initiate a reply message
Query reply message is back-propagated
File download
download A 7 1 4 2 6 3 5
Slide from Matei Ripenau;

21. Gnutella Maps
from: cf. figure 8-3 (p. 109)

22. A More Centralized Gnutella?
Reflectors (p. 112)
maintains an index of its neighbors
does not re-transmit the query, but answers from its own index
“mini-napster”
prelude to “super-nodes” (lecture on “performance”)
Host caches (p. 113)
bootstrapping your connection
cf. our list serve for
more convenient for users, but it doesn’t produce a nice random graph
everyone ends up in the tightly connected cell

23. Kan’s Take on Scaling Chapter 8 Horizon Cellular telephony analogy
number of nodes that can be “seen” within your TTL
only your neighbors really matter
Cellular telephony analogy
coverage radius
Ethernet
it works!

24. Ritter’s Take on Scaling
Jordan Ritter, “Why Gnutella Can’t Scale. No, Really”
description of Gnutella ca. early 2001
(go through the figures; not pasted in the slides)

25. Ripeanu, Iamnitchi & Foster’s Take on Scaling
“Mapping the Gnutella Network”, IEEE Internet Computing, 6(1), 2002
Gnutella continues to grow…
figure 1 from Ripeanu, Iamnitchi & Foster,
IEEE IC, 6(1), 2002

26. Sampling of Messaging figure 2 from Ripeanu, Iamnitchi & Foster,
IEEE IC, 6(1), 2002

27. Gnutella - Small World? figures 3&4 from Ripeanu, Iamnitchi & Foster,
IEEE IC, 6(1), 2002

28. Gnutella - Power Law? figures 5&6 from Ripeanu, Iamnitchi & Foster,
IEEE IC, 6(1), 2002

29. Gnutella Topology vs. Network Topology
figure 7 from Ripeanu, Iamnitchi & Foster,
IEEE IC, 6(1), 2002

30. Gnutella: 4-Cayley Trees
Gunther observes that Gnutella’s architecture is a 4-Cayley Tree
Gunther, “Hypernets - Good (G)news for Gnutella”
definition: “A tree in which each non-leaf graph vertex has a constant number of branches n is called an n-Cayley tree.”
from:
from Rains & Sloane, we find that Cayley worked on this ca
Rains & Sloane, “On Cayley's Enumeration of Alkanes (or 4-Valent Trees)”, Journal of Integer Sequences, Vol. 2 (1999), Article

31. Let’s Build a Gnutella Network!
N=2, TTL=2
N=2, TTL=5
N=5, TTL=5
N=5, TTL=2