Presentation is loading. Please wait.

Presentation is loading. Please wait.

SocioPlug Polystyrene: Survivable Shape for Self-Organising Data François Taïani Joint work with: Hoel Kervadec (INSA Rennes) Simon Bouget (ENS Rennes)

Similar presentations


Presentation on theme: "SocioPlug Polystyrene: Survivable Shape for Self-Organising Data François Taïani Joint work with: Hoel Kervadec (INSA Rennes) Simon Bouget (ENS Rennes)"— Presentation transcript:

1 SocioPlug Polystyrene: Survivable Shape for Self-Organising Data François Taïani Joint work with: Hoel Kervadec (INSA Rennes) Simon Bouget (ENS Rennes) Anne Marie Kermarrec (ASAP)

2 F. Taiani2 Focus Epidemic Topology Construction algorithms  Decentralized, fast, scalable  Fundamental building block to higher-level services (DHT, Multicast, Pub-Sub, Recommendations) Taken from [JMB09]

3 Problem: Catastrophic Failure The topology heals But the overall shape is lost How to recreate whole shape from surviving nodes? ?

4 Outline Background: Decentralized Topology Construction Polystyrene: Architecture and Protocol Evaluation Outlook F. Taiani4

5 Decentralized Topology Const. Each node : some data Find k “closest” nodes in system  Decentralized approach, asynchronous rounds random sampling (RPS) topology layer gossip- based topology construction (e.g. T-Man) topology linkrandom link A B C D E A B C D E nodenode position

6 Main idea: greedy neighbourhood optimization Decentralized Topology Const. exchange of neighbors lists neighborhood optimization 12 FC ED A B C D F E A B C D F E

7 Polystyrene Topology Construction (T-Man, Vicinity, Gossple) Neighbours Peer Sampling Service (RPS, Cyclon, SCAMP) Node position Polystyrene’s Architecture

8 Polystyrene Protocol recovery 2 FD Topology Construction migration 4 projection 3 Node position ghosts backup (outgoing) 1 backup (incoming) 1’ Neighbours 3’ guests

9 The Migration Process F. Taiani9 q.pos t p.pos t p.guests t q.guests t a b c d f e pq

10 The Migration Process Bi-clustering of guest points  Heuristics : diameter F. Taiani10 a b c d f e pq

11 The Migration Process Bi-clustering of guest points  Heuristics : diameter F. Taiani11 a b c d f e points closer to b points closer to d pq

12 The Migration Process Bi-clustering of guest points  Heuristics : diameter + minimum move F. Taiani12 q.pos t+1 p.pos t+1 q.guests t+1 a b c d f e p.guests t+1 points closer to b points closer to d pq

13 Evaluation Shape : 2D 40x80 logical torus Round 20 : 50% correlated node crashes F. Taiani13 Polystyrene recreates shape with surviving nodes After failure (r=20)(r=22)(r=28)

14 Eval: Quality of Neigborhoods Polystyrene maintains good neighborhoods

15 Eval: Quality of Shape And the torus gets restored!

16 Eval: Scalability F. Taiani16 Time (rounds) until homeogeneity less than Logarithmic convergence!

17 Outlook An example of advanced topology construction  Replicated, highly robust, self-organising Potential extension to load-balancing  Good for plug heterogeneity Concrete application on top of polystyrene  DHT, recommendation, queries, search Larger picture  Self-organising data primitives for plug infrastructures

18 References [KMG03] Kermarrec A.-M., Massoulie L., Ganesh, A.J., Reliable Probabilistic Communication in Large-Scale Information Dissemination Systems, IEEE Transactions on Parallel and Distributed Systems, March 2003, (14:3) [JGK04] Jelasity, M., Guerraoui, R., Kermarrec, A.-M., and van Steen, M. (2004). The peer sampling service: experimental evaluation of unstructured gossip- based implementations. Middleware ’04, pages 79–98, New York, NY, USA. Springer- Verlag New York, Inc. [VS05] Voulgaris, S. & Steen, M. V. Epidemic-style Management of Semantic Overlays for Content-Based Searching. Proc. of the 11th Int. Euro-Par Conf. on Parallel Processing (Euro-Par'05), Springer, 2005, 1143-1152 [Jelasity, Alberto Montresor, and Ozalp Babaoglu. 2009. T-Man: Gossip-based fast overlay topology construction. Comput. Netw. 53, 13 (August 2009), 2321- 2339. [BFG+10] Bertier, M.; Frey, D.; Guerraoui, R.; Kermarrec, A.-M. & Leroy, V. The GOSSPLE anonymous social network. Proc. of the ACM/IFIP/USENIX 11th Int. Conf. on Middleware, 2010, 191-211 [TLB14] Taiani, F., Lin, S. and Blair, G. S. (2014) GossipKit: A Unified Component Framework for Gossip. IEEE TSE, Preprint, doi: 10.1109/TSE.2013.50 18


Download ppt "SocioPlug Polystyrene: Survivable Shape for Self-Organising Data François Taïani Joint work with: Hoel Kervadec (INSA Rennes) Simon Bouget (ENS Rennes)"

Similar presentations


Ads by Google