A Computing Resource Discovery Mechanism Over a P2P Tree Topology - PowerPoint PPT Presentation

A Computing Resource Discovery Mechanism Over a P2P Tree Topology A Computing Resource Discovery Mechanism Over a P2P Tree Topology D. Castellà H. Blanco F . Giné F . Solsona Distributed Computing Group University of Lleida Spain VECPAR 2010 BERKELEY (CALIFORNIA)

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Introduction 1 System model 2 Rebalancing algorithm 3 Resource Discovery Service 4 Experimental results 5 Conclusions and future work 6

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Introduction Outline Introduction 1 System model 2 Rebalancing algorithm 3 Resource Discovery Service 4 Experimental results 5 Conclusions and future work 6

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Introduction Starting point Distributed computing 1000000000 100000000 10000000 1000000 100000 GigaFlops/s 10000 1000 100 10 1 Laptop Jaguar Internet Main goal To design and implement a P2P distributed computing system.

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Introduction Main functionalities of CoDiP2P Previous Work CodiP2P features: Connecting algorithm. 1 Updating algorithm. 2 Peer removal algorithm. 3 Job scheduling algorithm. 4

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Introduction Main functionalities of CoDiP2P Proposals Our proposal in this presentation: Rebalancing algorithm. 1 Searching algorithm. 2

A Computing Resource Discovery Mechanism Over a P2P Tree Topology System model Outline Introduction 1 System model 2 Rebalancing algorithm 3 Resource Discovery Service 4 Experimental results 5 Conclusions and future work 6

A Computing Resource Discovery Mechanism Over a P2P Tree Topology System model System model Miscellaneous Area of peers ( A i ). Peer Roles: Manager ( M j ) Worker ( W j ) Replicated Manager ( RM j )

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Rebalancing algorithm Outline Introduction 1 System model 2 Rebalancing algorithm 3 Resource Discovery Service 4 Experimental results 5 Conclusions and future work 6

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Rebalancing algorithm System restructuring Rebalancing algorithm Introduction Causes of having unbalanced tree: The churn of peers can unbalance the system: Massive input of peers into the system. Massive output of peers from the system. Solution Using a rebalancing algorithm for restructuring the system.

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Rebalancing algorithm System restructuring Rebalancing algorithm Three cases to consider: Case 1: Area A i isolated from its above manager Case 2: Area A i full with ( LPeer max − LPeer min ) > 1 Case 3: Area A i not full and LPeer max > 0 M1 M1 CASE 3 M3 M3 M4 M4 M2 M2 CASE 1 CASE 2 CASE 3 CASE 2 M5 Peer_max Peer_min Peer_max Peer_max Peer_min Peer_max W10 CASE 1 M5 W10 W8 W8

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Rebalancing algorithm System restructuring Rebalancing algorithm Cost of the algorithm: log | Area | ( N ) M1 M1 CASE 3 M3 M3 M4 M4 M2 M2 CASE 1 CASE 2 CASE 3 CASE 2 M5 W10 Peer_max Peer_min Peer_max Peer_max Peer_min Peer_max CASE 1 M5 W10 W8 W8

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Resource Discovery Service Outline Introduction 1 System model 2 Rebalancing algorithm 3 Resource Discovery Service 4 Experimental results 5 Conclusions and future work 6

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Resource Discovery Service Resource Discovery Service Overview In order to schedule and execute tasks: We have implemented a resource discovery service taking into account: A tree-like topology. 1 The information provided by the updating algorithm to the 2 managers. Able to perform exact and range queries. 3 Lookup CPU power available and returns the address of 4 peers owners.

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Resource Discovery Service Resource discovery service Searching algorithm: step 0

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Resource Discovery Service Resource discovery service Searching algorithm: step 1

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Resource Discovery Service Resource discovery service Searching algorithm: step 2

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Resource Discovery Service Resource discovery service Searching algorithm: step 3

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Resource Discovery Service Resource discovery service Searching algorithm: step 4

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Resource Discovery Service Resource discovery service Searching algorithm: Cost: log | Area | ( N )

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Experimental results Outline Introduction 1 System model 2 Rebalancing algorithm 3 Resource Discovery Service 4 Experimental results 5 Conclusions and future work 6

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Experimental results Experimental results Simulation Environment We use a GridSim based Simulator to design, implement and test all proposals. Number of nodes: 10000. Number of searches: 125.000 searches/s with a Poisson distribution with λ = 125 searches/s Network latency: 100ms. Network bandwidth: 1Mbps.

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Experimental results Experimental results Influence of the rebalancing algorithm over the searching algorithm 100 CoDiP2P 1 RM (With Rebal.) CoDiP2P 3 RM (With Rebal.) CoDiP2P 1 RM (No Rebal.) CoDiP2P 3 RM (No Rebal.) 80 % Unsuccessful Searches 60 40 20 0 0 10 20 30 40 50 60 % Failed Peers

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Experimental results Experimental results Reliability of exact queries searching algorithm 100 100 CoDiP2P 1 RM (Exact Query) CoDiP2P 1 RM (Exact Query) CoDiP2P 3 RM (Exact Query) CoDiP2P 3 RM (Exact Query) Chord 1 succ. (Exact Query) Chord 1 succ. (Exact Query) Chord 3 succ. (Exact Query) Chord 3 succ. (Exact Query) 80 80 % Unsuccessful Searches % Unsuccessful Searches 60 60 40 40 20 20 0 0 0 10 20 30 40 50 60 0 10 20 30 40 50 60 % Failed Peers % Failed Peers Freq = 125000 searches / s Freq = 125 searches / s

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Experimental results Experimental results Reliability of range queries searching algorithm 100 100 CoDiP2P 1 RM (Range Query) CoDiP2P 1 RM (Range Query) CoDiP2P 3 RM (Range Query) CoDiP2P 3 RM (Range Query) BATON (Range Query) BATON (Range Query) 80 80 % Unsuccessful Searches % Unsuccessful Searches 60 60 40 40 20 20 0 0 0 10 20 30 40 50 60 0 10 20 30 40 50 60 % Failed Peers % Failed Peers Freq = 125000 searches / s Freq = 125 searches / s

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Experimental results Experimental results System time response 100 Chord 1 succ. (Exact Query) CoDiP2P 1 RM (Exact Query) 80 % Unsuccessful Searches 60 40 20 0 0 500 1000 1500 2000 2500 3000 Seconds CoDiP2P and Chord comparative of system time response due to a 50 % of Failed Peers.

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Conclusions and future work Outline Introduction 1 System model 2 Rebalancing algorithm 3 Resource Discovery Service 4 Experimental results 5 Conclusions and future work 6

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Conclusions and future work Conclusions Achievements Over CoDiP2P system (tree topology): We have developed the main features: connection, updating, and peer removal algorithms. We have proposed two new algorithms: rebalancing and searching algorithms. The rebalancing algorithm ensures stability to the system. The reliability of lookups in CoDiP2P is higher than other topologies, specially in high frequencies. The system time response of CoDiP2P is lower than other topologies.

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Conclusions and future work Future work Avenues Testing and monitoring the tree topology and algorithms under real conditions and network. Extending the tree topology with a second level topology (Bruijn).

A Computing Resource Discovery Mechanism Over a P2P Tree Topology Conclusions and future work UdL VECPAR 2010