1. Information Federation in Grid Information Services Mehmet S. Aktas Advisor: Prof. Geoffrey C. Fox Ph.D. Defense Exam May 3, 2007

2. Talk Outline  Use Cases and Challenges  Research Issues  Architecture  Hybrid Grid Information Service  Performance Evaluation  Conclusions  Contributions and Future Research Directions 2

3. Introduction  Grid Information Services in Service Oriented Architectures  1) Large scale relatively static metadata as in catalog of all the world’s services Interaction-independent, slowly-varying metadata  2) Small scale highly dynamic metadata as in dynamic workflows for sensor integration and collaboration Interaction-dependent, dynamic metadata Dynamic Grid/Web Service Collections* – Dynamically assembled relatively small number of services (sub-grid) – Gathered at any one time to support a specific task – Generate dynamic metadata and have limited life-time [*] [ICCSE-05] Managing Dynamic Metadata as Context http://grids.ucs.indiana.edu/ptliupages/publications/maktas_iccse05.pdf 3

4. Motivating Use Cases  Geophysical Data Grids - CGL  Service Oriented Architecture for Geographical Information Systems Supporting Real Time Data Grids  Pattern Informatics (PI) - UC Davis  Earthquake forecasting code developed by Prof. John Rundle (UC Davis) and collaborators, uses seismic archives.  Interdependent Energy Infrastructure Simulation System (IEISS) - LANL  Models infrastructure networks (e.g. electric power systems and natural gas pipelines) and simulates their physical behavior, interdependencies between systems.  eSports System - CGL  Annotative collaboration application. Supports archive, replay, annotation of real-time video-conferencing streams. 4

5. A General Geographical Information System Grid Orchestration Scenario* [*] Building and Applying Geographical Information System Grids, Special Issue on Geographical information Systems and Grids based on GGF15 workshop, Concurrency and Computation: Practice and Experience http://grids.ucs.indiana.edu/ptliupages/publications/GISGrids_Concurrency_submitted.pdf 5

6. Background  Specifications for interaction-independent metadata  UDDI Specification  Glue Specification  EbXML Specification  Web Registry Service Specification  Specifications for interaction-dependent metadata  Point-to-point approach Web Service Resource Framework (WSRF) Specification  Third-party approach WS-Context Specification 6

7. Challenges  Standardization and Unification Issues  Customized Grid Information Services  Fat clients  Performance and Centralization Issues  Low performance  Low fault tolerance  UDDI Specification Issues  Lack of up-to-date, metadata-oriented registry  Lack of domain-specific metadata management  WS-Context Specification Issues  Limited data model and communication protocol 7

8. Research Issues I  Unification  How to combine different information services?  Federation  How to federate different information services?  Flexibility  How to accommodate broad range of specific application domains?  Interoperability  How to facilitate connection with wide range of information service clients? 8

9. Research Issues II  Performance  How to provide efficient information management strategies? high-performance, scalable in-memory storage efficient request distribution adaptation to instantaneous client-demand changes  Fault-tolerance  How to provide efficient replica-content placement strategies?  Consistency  How to provide efficient consistency enforcement strategies? 9

10. Hybrid Grid Information Service  Unification  Federation  Unified Schema  Query/Publish API  Flexibility  Interoperability  Extended UDDI  WS-Context  Glue  … Hybrid Grid Information Service  Unification  Federation  Unified Schema  Query/Publish API  Flexibility  Interoperability  Extended UDDI  WS-Context  Glue  … 10

11. UDDI instance WS-Context instance Unified schema instance 11

12. 12

13. Support for interaction-independent metadata: Extended UDDI Service  There are other extensions of UDDI  Supports different types of metadata  User-defined metadata  Functional metadata  Enables advanced query capabilities  Geo-spatial, metadata-oriented, domain-independent queries  Provides additional capabilities  Up-to-date service registry information (leasing)  Dynamic aggregation of capabilities of services e.g. geospatial capabilities [GGF16-Semantic Grid Workshop] Web Service Information Systems and Applications http://www.semanticgrid.org/OGF/ggf16/papers/GGF16SemGrid-CGL.pdf [SKG06 – IEEE Proceedings] XML Metadata Services http://grids.ucs.indiana.edu/ptliupages/publications/SKG2006_CameraReady_FinalFix.pdf 13

14. Support for interaction-dependent metadata: WS-Context Service  OASIS Standard  Context Manager Service  Data model and communication protocol  Supports Dynamic Web Service Collections  Distributed state based systems e.g. workflow-style grids  Session metadata management e.g. real-time replay and session-failure recovery capabilities  Provides various capabilities  Notification capability  Up-to-date metadata registry (leasing) [SKG05 – IEEE Proceedings] Information Services for Dynamically Assembled Semantic Grids http://grids.ucs.indiana.edu/ptliupages/publications/skg05-56-maktas-ieee-version.pdf [FGCS - 2007] Fault Tolerant High Performance Information Services for Dynamic Collections of Grid and Web Services http://www.informatik.uni-trier.de/~ley/db/journals/fgcs/fgcs23.html#AktasFP07 14

15. Support for federated service metadata: Information Federation  Federating Grid Information Services  Unified Schema and communication protocol  Extended UDDI, WS-Context and Glue Sche mas  Approach taken for Unified Schema [Schema Integration]  Schema Matching Identify overlapping information in given two Schemas: S1 and S2  Schema Merging Use the identified overlapping information to guide merge of S1 and S2  Communication protocol  Publish: save_ (create, update), delete_ e.g. save_service, delete_service  Inquiry: find_, get_ e.g. find_metadata, get_metadataDetail 15

16. Schema Matching: Identifying Matching Concepts serviceAttributeEntity: Information about metadata associated to services Site Service ComputingEl ement StorageElem ent site: information about a site where services, computing elements and storage elements are aggregated ServiceData serviceData: information associated to a service service: all information about a Service ExtUDDI.businessEntity 1:N GLUE.site ExtUDDI.businessService 1:1 GLUE.service ExtUDDI.serviceAttributeEntity 1:1 GLUE.serviceData Extended UDDIGLUE EXtUDDIGLUE 16

17. metadata: information about metadata associated to service bindingTemplate: Technical information about a service point tModel: Description of Specifications for services or taxonomies publisherAssertions: Defines relationships between two business entities computingElement: all info. required to manage computing resources storageElement: all information required to manage storage resources businessEntity: information about the party who publishes information about entities service: all information about a service site: all information about a concept to aggregate services and resources site contains one to n computing element has references to site contains one to n services site contains one to n storage element business contains one to n services has references to service contains one to n metadata service contains one to n technical information business contains one to n site Schema Merging: Unifying Schemas ExtUDDI.businessEntity  ExtUDDI&GLUE.businessEntity ExtUDDI&GLUE.site  GLUE.site ExtUDDI.businessService  ExtUDDI&GLUE.service  GLUE.service ExtUDDI.serviceAttribute  ExtUDDI.metadata  GLUE.serviceData Unified SchemaGLUEExtended UDDI Example Mappings => 17

18. Key Design Features  In-Memory storage  High performance metadata access/storage  Access distribution  Redirecting client request to an appropriate replica server  Replica content placement for performance  Dynamic replication Moving/replicating metadata to where they are demanded.  Replica content placement for fault-tolerance  Permanent replication Replicating data on an appropriate replica server  Consistency enforcement  Ensuring all replicas of a data to be the same 18

19. In-Memory Storage  Light-weight implementation of JavaSpaces  Data sharing, associative lookup  Integrated in-memory storage capability  Ex: UDDI-type, WS-Context-type  Today’s servers are capable of holding such small size metadata in memory.  Persistency  Newly-inserted/updated metadata is backed-up into appropriate information service back-end.  If the physical memory wiped out, at the bootstrap, database-metadata is inserted into the in-memory storage from the last-backup. 19

20. Experiment Results 20

21. Experiment Results 21

22. Message rate scalability investigation results 22

23. Message rate scalability investigation results 23

24. Access Distribution and Dynamic Replication  Broadcast-based request dissemination  Pub-sub system for message broadcast  Requests are broadcast only to those servers that can answer  No need to keep track of metadata locations  Replica-content placement for performance  Popular copies are moved/replicated where they are demanded  Dynamic migration/replication algorithm*  Self-adaptation to changing client demands [*] Rabinovich et al, A dynamic Object Replication and Migration Protocol for an Internet Hosting Service Proceedings of the 19th IEEE International Conference on Distributed Computing Systems, 1999 http://portal.acm.org/citation.cfm?coll=GUIDE&dl=GUIDE&id=880582 24

25. Access Distribution Experiment Benchmark Methodology T1T2T3 Time = T1 + T2 + T3 Simulation parameters Backup frequencyevery 10 seconds Message size2.7 Kbytes One-broker case Two-broker case 25

26. Experiment Results  Overhead of access distribution is only few milliseconds.  Continuous access distribution operation does not degrade the performance. 26

27. Experiment Results  The overhead of distribution remains the same regardless of the network distances between nodes. 27

28. T1T2T3 Time = T1 + T2 + T3 Dynamic Replication Performance Experiment Benchmark Methodology Simulation parameters message size / message rate2.7 Kbytes / 10 msg/sec replication decision frequencyevery 100 seconds deletion / replication threshold0.03 request/second and 0.18 request/second registry size1000 metadata in Indianapolis 28

29.  The decrease in average latency shows that the algorithm manages to move replica copies to where they are demanded. Experiment Results 29

30. Replication and Consistency  Permanent replication for fault tolerance  Each node keeps information about other servers  Replica Server(s) Selection Load and proximity metrics Selection algorithm by Rabinovich et al  Unicast-based replica-content placement  Primary-copy approach  Updates are unicast to primary-copy  Updates are broadcast by the primary-copy holder to a) permanent-copy holding servers b) applications with high consistency requirements 30

31. Fault-tolerance Experiment Benchmark Methodology T1T2T3 Time = T1 + T2 + T3 Simulation parameters Backup frequencyevery 10 seconds Message size2.7 Kbytes One-broker case Two-broker case 31

32. Experiment Results  Overhead of replica-content placement is only few milliseconds.  Overhead of replica-content placement increases in the order of milliseconds as the fault-tolerance level increase. 32

33. Consistency Enforcement Experiment Benchmark Methodology T1T2T3 Time = T1 + T2 + T3 Simulation parameters Backup frequencyevery 10 seconds Message size2.7 Kbytes One-broker case Two-broker case 33

34. Experiment Results  Overhead of consistency enforcement is few milliseconds.  The cost of consistency enforcement remains the same regardless of distribution of the network nodes. 34

35. Contributions  Systems Research  Hybrid Grid Information Service Architecture  Unification, Federation and Interoperability of grid information services  Strategies for high-performance, scalable in-memory storage  Strategies for efficient distribution, replica-content placement, consistency enforcement by utilizing pub-sub based messaging schemes  Self-adaptation to changing-client demands  Extensions to semantics of UDDI and WS-Context Web Service Specifications  Detailed evaluation of the system components and algorithms  Systems Software  An implementation of Extended UDDI Specification  Geographical Information Systems-specific, metadata-oriented  An implementation of WS-Context Specification  Session metadata management for collaboration grids, distributed state management for workflow-style grids  An implementation of Hybrid Grid Information Service Architecture 35

36. Future Research Directions  Use the proposed approach to solve OGF Grid Interoperation Now (GIN) problem for information services  Investigate an information security mechanism for the decentralized Hybrid Service  Example motivating application case: Pattern Informatics application  Applying Hybrid Service to broader range of application use cases  Web 2.0/Folksonomy information services 36