Grid Quality of Service and Service Level Agreements

Grid Quality of Service and Service Level Agreements Karim Djemame University of Leeds

Outline • • • SLA: What is in it? SLA Management Architecture Example: SLA for a Compute Service Need for Resource Brokering Conclusion and Future Work Distributed Aircraft Maintenance Environment - DAME

Service Level Agreements (SLA) • • • A Service Level Agreement is • “…an explicit statement of expectations and obligations that exist in a business relationship between two organisations: the service provider and the customer. ” Formalises requirements and specifies behaviours A description of the SLA Model: service being provided SLA Root Purpose Names of the provider, consumer and 3 rd Parties Provider Consumer Scope SLIndicator SLObjectives Levels of service (guarantees) agreed on 3 rd Parties A definition of the domains in which the SLA is valid Measurement State Violations Exclusions Policing What is not covered by the SLA Describes the SLA state Distributed Aircraft Maintenance Environment - DAME

DAME Problem Scenario User wants access to computational resources on the Grid GSx HTTPS Globus GT 3 d. B GSx HPCx Internet HTTPS GSx Resources Portal access over internet Consumer Specialist Provider Resources + Grid Services Grid resource access via Grid Services and the Globus Toolkit White Rose Grid Distributed Aircraft Maintenance Environment - DAME

Proposed SLA Management Architecture Provides access to service functions Provides meta-data relating to the service eg. state Selects resources and submits the execution Reacts to violations by adapting managed grid service execution Negotiates SLA between the user and provider Monitors Service Level Objectives and records violations Distributed Aircraft Maintenance Environment - DAME

SLA Management Interaction Capture user requirements Interface Negotiate SLA Sign SLA Launch managed grid service Reserve resources Monitor and record violations Adapt to violations Distributed Aircraft Maintenance Environment - DAME

SLA Management • • Automated management will provide autonomic self-optimisation without the need for user intervention Outcome: for each Grid Execution instance, an SLA Manager will provide: – SLA – resource reservation – job monitoring – job adaptation – violation log Distributed Aircraft Maintenance Environment - DAME

Example: SLA for a Compute Service Purpose Parties Scope Run a Grid job with guarantees Ensure the user’s requirements are met The user, the broker, the resources Compute service Service Level Objectives Availability of resources Attributes: CPU count, CPU type, CPU speed, RAM, Storage, Operating system and version Service Level Indicators Resource availability and value of each attribute Exclusions What is not currently supported Administration SLA’s objectives met through resource brokering Distributed Aircraft Maintenance Environment - DAME

SLA Specification using XML • Grid Task requirements represented in an SLA content tree (JAVA) • SLA content tree converted to XML using Java Architecture for XML Binding (JAXB) • SLA XML document created to interact with external resource brokers and pass user requirements Distributed Aircraft Maintenance Environment - DAME

SNAP Resource Broker Architecture TSLA Portal TSLA: Task Service Level Agreement RSLA: Resource Service Level Agreement BSLA: Bind Service Level Agreement Matchmaker Match. Maker RSLA Decision Maker Knowledge bank Co-Ordinator Co-ordinater Decision maker Dispatcher Resource gatherer Resource Gatherer BSLA Grid Middleware Grid middleware Grid Resources R R SNAP: Service Negotiation and Acquisition Protocol Distributed Aircraft Maintenance Environment - DAME

Current and Future Work • • • Implementation • • Produced a basic implementation of an SLA Manager to accept user requirements, embed them within an SLA (XML) document and report them back to the user SNAP-based resource broker being deployed on the WRG Automated Monitoring • • Engineering automated monitoring within the SLA Manager Enable subscriptions to an external Grid Monitoring Service depending on the SLA guarantees Adaptation Algorithm • • Investigate an adaptation algorithm that is suitable for use within a Service Oriented Architecture Integration with SLA Manager Distributed Aircraft Maintenance Environment - DAME

References • SLA Management in a Service Oriented Architecture. K. Djemame, M. Haji and J. Padgett. ICCSA’ 2004, Assissi, Italy, May 2004 • A SNAP-based Community Resource Broker using a Three-Phase Commit Protocol. M. Haji, P. Dew, K. Djemame and I. Gourlay. IPDPS’ 2004, Santa Fe, New Mexico, April 2004 Distributed Aircraft Maintenance Environment - DAME