Transcript social network

Mao Yang, HuaZhang
Chen, Ben Y.
Zhao, Yafei Dai, and Zheng
Zhang
P2P
file sharing
 Popular




techniques
NAPSTER
Kazaa
Overnet and
BitTorrent





Napster used centralized servers to index available files on its
application nodes.
Kazaa uses a two-level hierarchical structure, where
supernodes stores indices of files shared by nearby clients.
Mojonation used a virtual currency (mojos) to provide
incentive for cooperative sharing.
BitTorrent enforces a modified version of the pair-wise titfor-tat data sharing model for clients performing simultaneous
downloads.
BitTorrent targets clients who are actively downloading the
same document.
Scalability and performance
 Operation and the impact incentives and
mechanisms on
 User behavior
 Highly distributed indexing







CERNET: China Education and Research Network.
Provide a large amount of publicly accessible software and
documents to educational computing users in China.
FTP servers across the high-bandwidth CERNET server the
files to the users.
T-Net (search engine) addresses the problem of locating
documents across these servers.
T-Net did not solve the basic problems of FTP servers: limited
bandwidth and availability.
User feed back driven incentive policies.

Operations in Maze:
 Clients upload file metadata to a Maze
server
 Metadata is replicated to a subset of index
servers, where they are indexed
 Clients send queries to Maze search
servers and
 Queries are resolved by index servers




The network should locate replicas in nearby networks
whenever possible for efficiency.
Reduce the occurrence of “free-riding”(log off after
downloads) which requires a strong incentives
mechanism that encourages users to share their
resources.
Maze should leverage social relationships between
users to improve efficiency of searches and quality of
results.
Retain full control over code and deployment to
leverage it as platform to experiment with different
designs, incentive and security policies, and as a source
of detailed file-sharing measurements.
1. Maze’s operation is similar
to that of the Napster filesharing network.
2. Each file has the following
attributes
(Owner
ID,
Filename, Filetype, Size,
Creation date, MD5).
3. User can perform search on
any or a combination of
these attributes.
4. Several Index Servers store
information about all files
available on peer nodes
5. A
client periodically or
when comes online sends an
“ALIVE ” or “HEART
BEAT” signal to the heart
beat servers along with an
update of which files they
have currently available by
signature.
6. These are compared to those
stored on index servers, and
additional metadata is sent
for newly acquired files
7. When a client makes a
request, it goes to all the
index servers.
8. Which in-turn returns the
list of (online) nodes where
the file is available.
9. Preference is given to the
nodes with same location or
the same class of Ip
addresses.
10. The client with this list
triggers
a
“swarm
download”.
11. Maze
adds NAT-traversal
mechanisms to allow users
behind firewalls and NAT
boxes to communicate by
forwarding through their
peers.
12. Similar to Gnutella, Users
also build their own “social
network” by adding peers
to their friend lists.




How do you make users upload in a swarm ?
Or
How to avoid free-riding in a swarm ?
Maze relies on a set of incentive policies driven by
direct user feedback from public forums such as BBSes
Existing Approaches : • Game theory.
• Tit-for –Tat.
• Samsara
• Prop-share.

1.
2.
3.
MAZE uses a novel incentive mechanism as discussed : New users are initialized with 4096 points.
Uploads: +1.5 points per MB uploaded
Downloads:
Download range (MB)
Points (Per MB )
0 – 100
-1
100-400
-0.7
400-800
-0.4
800 - above
-0.1
4. Downloads requests are ordered by: T = requestTime− 3 ∗
logP, where P is a user’s point total.
5. Users with P < 512 have a download bandwidth quota of
300Kb/s.


A Bulletin Board System, is a
computer system running software
that allows users to connect and
log in to the system using a
terminal program.
Once logged in, a user can
perform functions such as
uploading and downloading
software and data, reading news
and bulletins, and exchanging
messages with other users, either
through electronic mail or in
public message boards.
Users used a number of ways to improve their points
level in the Maze system
 Ran multiple instances of Maze on a single machine,
and transferred files between them to artificially boost
their point scores.
 Switching to new identities when the current point level
has dropped significantly following downloads.
 Spoof new files by modifications made in the metadata
to highly popular files to boost their points.
 Embed popular search strings in the metadata to make
them appear more in other users' search results.





Maze is one of the first large-scale deployments of an
academic research project,
Maze has over 210,000 registered users and more than
10,000 users online at any time, sharing over 140
million files.
As of July 2004, Maze includes a user population of
210,000 users and supports searches on 140 million
files (20 million unique) totaling over 226TB of data.
At any given time, there are over 10,000 users online,
and over 2700 active searches or transfers occurring
simultaneously.




Maze uses an explicit IP address scoping mechanism to
provide locality-aware search results to the end user, resulting
in faster downloads and lower bandwidth consumption.
Maze supplements the normal network structure with a social
network, and uses incentives as a key part of its resource
allocation and download scheduling policies.
Maze uses a community discussion board to actively solicit
feedback on the incentive structure from the user population.
Maze relies on a set of incentive policies driven by direct user
feedback from public forums such as BBSes.(Bulletin board
systems)



In Maze, in addition to server based file indices, peers
connect to each other using a social network, and can
rely on friends to resolve queries and forward traffic
between hosts behind NAT boxes.
Maze uses IP address matching to recognize network
locality and encourage downloads from nearby
replicas.
Maze leverages an active user forum to determine and
encourage the use of an incentive policy.