proposal and evaluation of a cooperative mechanism for hybrid p2p …€¦ · proposal and...
TRANSCRIPT
PROPOSAL AND EVALUATION OF A COOPERATIVE MECHANISM FOR
HYBRID P2P FILE-SHARING NETWORKS
Hongye Fu Naoki Wakamiya Masayuki MurataGraduate School of Information Science and Technology
Osaka [email protected]
2005/11/1 2
CIIT 2005
Content
• Backgrounds• Cooperation Mechanisms for a SPB Approach• Simulation Results• Conclusion and Future Works
2005/11/1 3
CIIT 2005
Competing Overlay Networks
• Overlay networks are widely deployed over physical IP networks to obtain application-oriented QoS
• Selfish overlay networks compete for limited physical resources and disrupt each other
2005/11/1 4
CIIT 2005
Overlay Network Symbiosis
• Overlay networks improve and enhance their collective performance by cooperation
• Overlay networks = Organisms– Evolve as a new node joins– Shrink as a node leaves– Direct or indirect interactions– Change internal structures
Symbiotic colony of E. coli and Dictyostelium
2005/11/1 5
CIIT 2005
Inter-Overlay Communications
• Each node in an overlay network autonomously establish or terminate logical links– within an overlay network– with another overlay network
• Message exchanges over a logical link• A logical link is kept as far as the both sides benefit
from the link (mutualism)• Overlay networks come to merge together
inter-overlay links
2005/11/1 6
CIIT 2005
A Cooperative Mechanism for P2P networks
• In this work, we focus on the cooperation among hybrid P2P file-sharing networks (Napster, winMX...)
• P2P networks exchange query and response messages with each other
• Benefits– A peer can find more files at more peers– A peer can choose the best, i.e., the fastest or the
most reliable peer among many provider peers found in a search
– The stability of the whole system will be improved
2005/11/1 7
CIIT 2005
Hybrid P2P File-sharing Networks
1. Query
2. F
orwa
rd2. Forward
2. Forward 3. R
espo
nse
3. Response
4. R
eque
st
5. T
rans
mit
The process of retrieving a file～ The case that Peer A retrieves a file ～0. A peer connects with a meta-server and
registers meta-information about files to share with the other peers
1. Peer A sends a query message2. The meta-server forwards the query
message to other neighboring meta-servers
3. The meta-server sends a response message to the querying peer when meta-information about the desired file exists in its directory
4. Peer A directly request a file from a provider peer (Peer B)
5. Peer B transmit a file to Peer A directly
Peer B
Peer CPeer DPeer E
Peer F
Peer G
Peer H
Peer I
Peer A
Meta-Server X
Meta-Server Y
Meta-Server Z
Logical Link
2005/11/1 8
CIIT 2005
Basic Mechanisms
• A node introduces a cooperative program to enhance its own application-level QoS
• A cooperative program:– discovers other P2P networks– decides whether P2P networks cooperate with each
other– cooperate by exchanging messages
• Two types of mechanisms– Shared-Peer-Based (SPB) approach– Server-Chain-Based (SCB) approach
2005/11/1 9
CIIT 2005
Cooperation Mechanisms for a SPB Approach
network1 network2
Query
Response
Our Goal：The cooperation is achieved in a transparent way where other meta-servers and peers are unaware of the cooperation
Problem: A meta-server does not forward a query message to a general peer
A cooperative program includes meta-server modulesA Meta-server module can
relay query and response messagesgenerate response messagesdeposit meta-information into its local cache
By introducing a cooperative program, a shared peer becomes a cooperative peer
A cooperative peer behaves not only as a peer, but also as a meta-server
2005/11/1 10
CIIT 2005
A querying peer retrieves a file through the cooperative peer
File Retrieval
1. Query
network1 network2
2. Forward
2. F
orw
ard
2. Forward 2. Forward2. F
orward
3. Response3. R
esponse
3. Response
5. Transmit
4. Request
Peer P
Peer Q
3. Response
4. Request5. Transm
it5.
Tra
nsm
it4.
Req
uest
Protocol conversion if needed
Finds a provider peer Q
Protocol conversion if neededDeposits meta-information
2005/11/1 11
CIIT 2005
Simulation Experiments
• No caching meta-information• F kinds of files are available
– The popularity of files follows a Zipf distribution with α = 1.0
– Files are assigned to peers • Peers generate query messages
following the poisson process– File to find are determined by
its popularity
An Example for Hybrid P2P File-Sharing Network Topology (m=5, n=100)
2005/11/1 12
CIIT 2005
Measures
• Application-level measures– Ratio of Available Files
– Hit Rate
• System-level measures– Load on Cooperative Peers– Load on Meta-Servers
Number of kinds of available files in a networkNumber of kinds of available files in two network (F)
Number of successful query messagesTotal number of query messages generated
2005/11/1 13
CIIT 2005
Simulation Results- Ratio of Available Files and Hit Rate -
• No-Cooperation– Peers can only find files in their own network
• Ratio of Available Files: 69 ~ 70%• Hit Rate: 89 ~ 95%
• Cooperation– Peers can find files
not only in their own network but also in the other• Ratio of Available Files: 100%• Hit Rate: 100%
• Ratio of Available Files increases by about 30%• Hit Rate also increases regardless of the network size and the degree of increase is higher with smaller networks
2005/11/1 14
CIIT 2005
Simulation Results- Influence of Number of CP on Load（1000:1000） -
0
500
1000
1500
2000
2500
3000
3500
4000
4500
0 1 2 3 4 5 6 7 8 9 10
Ave
rage
num
ber o
f mes
sage
s
Number of cooperative peers
meta-server (m=2)
meta-server (m=5)
cooperative peer (m=2)
cooperative peer (m=5)
meta-server (m=2)meta-server (m=5)coop-peer (m=2)coop-peer (m=5)
• the load on meta-servers almost linearly increases with the increase of the number of cooperative peers
• the increase in the load on meta-servers by cooperation is almost the same regardless of the number of meta-servers
the load on cooperative peers does not changemuch
2005/11/1 15
CIIT 2005
From a view point of the load on cooperative peers, which is usually less powerful than meta-servers, the cooperation among P2P networks with a small number of meta-servers is desirable
Simulation Results- Influence of Number of MS on Load （1000:1000） -
0
1000
2000
3000
4000
5000
6000
1 2 3 4 5 6 7 8 9 10
Ave
rage
num
ber o
f mes
sage
s
Number of meta-servers
meta-server (c=4)meta-server (no-coop)
cooperative peer (c=4)
meta-server (c=0)meta-server (c=4)
cooperative peer (c=4)
the load on meta-servers decreases with the increase of the number of meta-servers and it is the same for both in cooperative and non-cooperative networks
the load on cooperative peers increases as the number ofmeta-servers increases
2005/11/1 16
CIIT 2005
Conclusion and Future Works
• Conclusion– Two cooperative approaches are proposed for hybrid P2P file-
sharing networks to efficiently cooperate with each other to improve their collective application-level QoS
– Through simulation experiments• our cooperative mechanisms (SPB approach) can improve the
application-level QoS at the sacrifice of the increased load• the influence of network configurations (the number of peers and
meta-servers) is investigated
• Future work– An efficient cache algorithm for cooperative peers– A decision algorithm of cooperation– Cooperative mechanisms which take into account characteristics
of physical networks
2005/11/1 17
CIIT 2005
Thank you
• Questions?
2005/11/1 18
CIIT 2005
Reachability
• Maximum interconnection leads to higher reachability.
connected by random nodes
connected by high-degree nodes
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1 2 3 4 5 6 7 8 9 10
reac
habi
lity
TTL
10000100-connect50-connect10-connect5-connect1-connect
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1 2 3 4 5 6 7 8 9 10
reac
habi
lity
TTL
10000100-connect50-connect10-connect5-connect1-connect
2005/11/1 19
CIIT 2005
Connectivity
• Maximum interconnection is more robust against random removal, but it is fragile under intentional attacks.
connected by random nodes
connected by highest-degree nodes
0
0.2
0.4
0.6
0.8
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
conn
ectiv
ity
fraction of removed nodes
error
attack
10000100-connect50-connect10-connect5-connect1-connect
0
0.2
0.4
0.6
0.8
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
conn
ectiv
ity
fraction of removed nodes
error
attack
error
attack
error
attack
error
attack
10000100-connect50-connect10-connect5-connect1-connect