Upload
cain-allen
View
32
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Myconet: A Fungi-Inspired Model for P2P Superpeer Overlay Topologies. Paul Snyder, Rachel Greenstadt, and Giuseppe Valetto {pls29,greenie,valetto}@cs.drexel.edu Department of Computer Science Drexel University. Outline. Overview Protocol description Evaluation Conclusion. 2. - PowerPoint PPT Presentation
Citation preview
1
Myconet: A Fungi-Inspired Model forP2P Superpeer Overlay Topologies
Paul Snyder, Rachel Greenstadt, and Giuseppe Valetto
{pls29,greenie,valetto}@cs.drexel.edu
Department of Computer ScienceDrexel University
Outline
• Overview• Protocol description• Evaluation• Conclusion
2
3
Myconet: Self-organization of superpeer overlay topologies
• Self-organizing unstructured P2P overlay– Hierarchical superpeers
• Inspired by hyphae, the robust, root-like structures of fungal mycelia
• Goals– Effective exploitation of peers– Resilience to failures
Photo credit: K. Fleming.Reproduced under a Creative Commons license.http://www.flickr.com/photos/myriorama/101120710/
Quick Definitions
• Peer-to-peer• Overlay networks• Superpeers• Unstructured P2P
Myconet’s Metaphor
Mycelium
Hyphae
Nutrients/Biomass
Myconet overlay
Superpeers
Regular peers
Goals of the Myconet Protocol
• Peers operate with local information only• Use multiple protocol states to balance
exploration and exploitation• Select highest-capacity nodes as
superpeers• Quick recovery after node failure
6
Outline
• Overview• Protocol description• Evaluation• Conclusion
7
Myconet Basics
• Round-based simulation in PeerSim• Peers characterized by an integer capacity• Peers with direct hyphal links are
considered neighbors• Uses a lower-level overlay to
communicate node status information– The gossip-based Newscast protocol
3. loses parent hypha
1. s
pore
s
START
attachedbiomass
unattachedbiomass
extendinghypha
2. finds hypha
• All peers begin as disconnected biomass
• Peers that cannot find a hypha to connect to become extending hyphae (superpeers)
• Extending hyphae seek biomass and to connect to another hyphal peer
Protocol: Extending Hyphae
Protocol: Bootstrapping
10
Biomass peersExtending hyphae
Round 0
Protocol: Bootstrapping
11
Biomass peersExtending hyphae
Round 1
Protocol: Branching Hyphae• Extending hyphae with
enough biomass promote to branching hyphae, which:– Form inter-hyphal
connections– Absorb biomass from
extending peers– Regulate number of
extending peers in the network
– Demote to extending status if unable to maintain biomass
3. loses parent hypha
1. s
pore
s
5. reaches or
exceeds capacity
6. absorbed
7. falls below
full utilization
7. promoted by a branching
or immobile hypha
START
attachedbiomass
unattachedbiomass
branchinghypha
extendinghypha
2. finds hypha
12
Protocol: Superpeer Promotion
1313
Biomass peersExtending hyphaeBranching hyphae
Round 1
Protocol: Superpeer Promotion
1414
Biomass peersExtending hyphaeBranching hyphae
Round 2
Protocol: Superpeer Promotion
1515
Biomass peersExtending hyphaeBranching hyphae
Round 4
16
immobilehypha
3. loses parent hypha
1. s
pore
s
5. reaches or
exceeds capacity
6. absorbed
8. absorbed
7. falls below
full utilization
7. promoted by a branching
or immobile hypha
10. falls belowutilization threshold
9. reaches orexceeds target
hyphal link count
START
11.
abso
rbed
attachedbiomass
unattachedbiomass
branchinghypha
extendinghypha
2. finds hypha
• Branching hypha with Cn hyphal connections become immobile hyphae– In the network long
enough to be considered stable
– Pull biomass from extending and branching hyphae
– Maintain hyphal connections if lost
– Regulate growth or collapse of extending and branching hyphae
– Demote if unable to maintain utilization threshholds
Protocol: Immobile Hyphae
Protocol: Stability
17
Biomass peersExtending hyphaeBranching hyphaeImmobile hyphae
Round 4
Protocol: Stability
18
Biomass peersExtending hyphaeBranching hyphaeImmobile hyphae
Round 17
Protocol: Stability
19
Biomass peersExtending hyphaeBranching hyphaeImmobile hyphae
Round 22
Outline
• Overview• Protocol description• Evaluation• Conclusion
20
21
Evaluation
1. Convergence to stable configuration
2. Resource utilization
3. Approximation of optimal superpeer configuration
4. Resilience to catastrophic failures
Evaluation: Methodology
• Tested using round-based simulation in PeerSim• Most graphs represent experimental results for networks
of 105 nodes, averaged over 25 runs– Experiments were conducted with network sizes from 103 to 106
nodes• Peer capacities were assigned using a power-law
distribution– For networks of size of 105, the probability of peer n having
capacity x is P[cn = x] = x-2, with x in the interval [1,500]– Maximum capacities were adjusted for other network sizes– Also tested with uniform random distributions, with similar results
• Compared performance to SG-1 and ERASP
23
Evaluation: Time to Stability
For 105 nodes, Myconet quickly converges to around 225 superpeers
24
Evaluation: Utilization Levels
Within 20 rounds, 95% of peers are connected to a branching or immobile hypha
25
Evaluation: Superpeer Configuration
The total superpeer count closely tracks the theoretical optimum
26
Evaluation: Failure Recovery
Biomass peersExtending hyphaeBranching hyphaeImmobile hyphae
Round 22
Evaluation: Failure Recovery
27
Biomass peersExtending hyphaeBranching hyphaeImmobile hyphae
Round 39
Evaluation: Failure Recovery
28
Biomass peersExtending hyphaeBranching hyphaeImmobile hyphae
Round 40
29
Evaluation: Failure Recovery
Biomass peersExtending hyphaeBranching hyphaeImmobile hyphae
Round 57
Evaluation: Failure Recovery
30
Biomass peersExtending hyphaeBranching hyphaeImmobile hyphae
Round 71
31
Evaluation: Failure Recovery
The Myconet overlay quickly repairs itself after a catastrophic failure
32
Evaluation: Bottom Line
• Myconet effectively constructs and maintains a strongly-interconnected, decentralized superpeer overlay
• Quickly converges to an optimal number of superpeers and high levels of capacity utilization.
• Performance scales smoothly up to at least 106 peers• Compared to other the state of the art, our simulations
show Myconet fares well in terms of:– Network stabilization– Response to catastrophic failure– Capacity utilization
Outline
• Overview• Protocol description• Evaluation• Conclusion
33
Conclusions
• Myconet overlay demonstrates applicability of fungal metaphor to peer-to-peer overlays
• Take-aways:– Hierarchical superpeer states were useful in
designing self-organizing network dynamics– Choosing to underutilize some superpeers (extending
hyphae) useful when balancing exploration vs. exploitation
– Increasing Cn parameter increased instant resistance to disconnection, but had unexpectedly slight effects on dynamics
Future Work
• Examine performance under network churn• Measure overlay maintenance costs• Test the performance of P2P applications
running on the Myconet overlay• Dynamic adaptation of Cn parameter• Move from round-based simulation to
protocol implementation• Explore possibility of formalizing metaphor in
terms of more rigorous biological models
Questions?
Paul Snyder, Rachel Greenstadt, and Giuseppe Valetto{pls29,greenie,valetto}@cs.drexel.edu
37
References
[1] A. Montresor, “A robust protocol for building superpeer overlay topologies” in Proceedings of the 4th International Conference on Peer-to-Peer computing. Zurich, Switzerland: IEEE, Aug. 2004, CONFERENCE, pp. 202-209.
[2] W. Liu, J. Yu, J. Song, X. Lan, and B. Cao, “ERASP: An Efficient and Robust Adaptive Superpeer Overlay Network,” Lecture Notes in Computer Science, vol. 4976, p.468, 2008.