Resource Discovery & Matchmaking
Main Classes of Resource Discovery  [1] Centralized third party Single server provides info of resources Example: DNS, LDAP, Napster, Globus Distributed third party Replication of multiple servers to provide info Example: UDDI, NetSolve, CORBA Multicast Limitation of LAN boundaries Example: Jini, Salutation, UPnP, Ninja P2P Genuinely distributed system Example: Chord, CAN, Gnutella, Pastry, Tapestry
Convergence of P2P + Grid   [2] Grid  Nature: complexity (variety of resources + applications), administrative management + policy-based Requirement: scalability, intermittent participations P2P Nature: specific applications (either file sharing or CPU cycles), anonymity, intermittent participants Requirement: complexity and attributed-based search Grid + P2P Massively scalable sharing environment
Two Main Classes of P2P Unstructured P2P Flooding / forwarding queries Example: Gnutella, Freenet, BitTorrent Problem: Nondeterministic search, N/W traffic Structured P2P Distributed Hash Table (DHT) Example: Chord, CAN, Tapestry, Pastry, P-Grid Problem: must be exact-key search, complex algo
Iamnichi and Foster ‘s  [2] Architectural Components:  Membership protocol Overlay Construction Preprocessing Request Processing Emulated Grid: unstructured P2P of 32,768 virtual nodes Evaluated Resource Discovery Algorithms Random Walk: worst response but no cache Learning-based: best but require cache Best-neighbor: good for many distinct requests Learning-based + best-neighbor: unpredictable
Sivadon and Putchong ’s   [3] Unstructured Hierarchical P2P Form top-most  to lowest: Global, VO, Super, Edge Flooding-based query algorithms Query-filter Backing links Backing resource links Simulation   (implemented by Sugree’ Hypersim) 100,000 peers located in 8 VOs Evaluated algorithms by determining swamping problem + response time
Condor’s Matchmaking  [4]   Condor’s Matchmaking Most distributed systems are basing on name/keyword search Matchmaking’s idea: advertise resources via classads Classads = advertisement + query Two Phases of Matchmaking Matching “ rank” = preference “ constraint” = requirement Claiming Try each one in the list of matching resources e.g., security issue + support different allocation models
References K. Vanthournout, G. Deconinck, and R. Belmans, “A Taxonomy for Resource Discovery, Personal and Ubiquitous Computing”, Springer Verlag London, UK, Volume 9, Issue 2, 2005 A. Iamnitchi and I. Foster, “A Peer-to-Peer Approach to Resource Location in Grid Environments”, Grid Resource Management, Kluwer Publishing, 2003 S. Chaisiri and P. Uthayopas, “Performance Evaluation of Peer-to-Peer Approach to Resource Discovery for Large Scale Grid Environments”, Proceedings of The 8th Annual National Symposium on Computational Science and Engineering, Thailand, 2004 R. Raman and M. Livny, “Matchmaking: Distributed Resource Management for High Throughput Computing”, Proceedings of the 17th IEEE International Symposium on High Performance Distributed Computing, 1998

Research Issues on Resource Discovery & Matching Making

  • 1.
  • 2.
    Main Classes ofResource Discovery [1] Centralized third party Single server provides info of resources Example: DNS, LDAP, Napster, Globus Distributed third party Replication of multiple servers to provide info Example: UDDI, NetSolve, CORBA Multicast Limitation of LAN boundaries Example: Jini, Salutation, UPnP, Ninja P2P Genuinely distributed system Example: Chord, CAN, Gnutella, Pastry, Tapestry
  • 3.
    Convergence of P2P+ Grid [2] Grid Nature: complexity (variety of resources + applications), administrative management + policy-based Requirement: scalability, intermittent participations P2P Nature: specific applications (either file sharing or CPU cycles), anonymity, intermittent participants Requirement: complexity and attributed-based search Grid + P2P Massively scalable sharing environment
  • 4.
    Two Main Classesof P2P Unstructured P2P Flooding / forwarding queries Example: Gnutella, Freenet, BitTorrent Problem: Nondeterministic search, N/W traffic Structured P2P Distributed Hash Table (DHT) Example: Chord, CAN, Tapestry, Pastry, P-Grid Problem: must be exact-key search, complex algo
  • 5.
    Iamnichi and Foster‘s [2] Architectural Components: Membership protocol Overlay Construction Preprocessing Request Processing Emulated Grid: unstructured P2P of 32,768 virtual nodes Evaluated Resource Discovery Algorithms Random Walk: worst response but no cache Learning-based: best but require cache Best-neighbor: good for many distinct requests Learning-based + best-neighbor: unpredictable
  • 6.
    Sivadon and Putchong’s [3] Unstructured Hierarchical P2P Form top-most to lowest: Global, VO, Super, Edge Flooding-based query algorithms Query-filter Backing links Backing resource links Simulation (implemented by Sugree’ Hypersim) 100,000 peers located in 8 VOs Evaluated algorithms by determining swamping problem + response time
  • 7.
    Condor’s Matchmaking [4] Condor’s Matchmaking Most distributed systems are basing on name/keyword search Matchmaking’s idea: advertise resources via classads Classads = advertisement + query Two Phases of Matchmaking Matching “ rank” = preference “ constraint” = requirement Claiming Try each one in the list of matching resources e.g., security issue + support different allocation models
  • 8.
    References K. Vanthournout,G. Deconinck, and R. Belmans, “A Taxonomy for Resource Discovery, Personal and Ubiquitous Computing”, Springer Verlag London, UK, Volume 9, Issue 2, 2005 A. Iamnitchi and I. Foster, “A Peer-to-Peer Approach to Resource Location in Grid Environments”, Grid Resource Management, Kluwer Publishing, 2003 S. Chaisiri and P. Uthayopas, “Performance Evaluation of Peer-to-Peer Approach to Resource Discovery for Large Scale Grid Environments”, Proceedings of The 8th Annual National Symposium on Computational Science and Engineering, Thailand, 2004 R. Raman and M. Livny, “Matchmaking: Distributed Resource Management for High Throughput Computing”, Proceedings of the 17th IEEE International Symposium on High Performance Distributed Computing, 1998