1 / 18

Towards GLUE Schema 2.0

This document provides modeling guidelines for using GLUE Schema 2.0, including its relationship with JSDL and CIM.

asteward
Télécharger la présentation

Towards GLUE Schema 2.0

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Towards GLUE Schema 2.0 Sergio Andreozzi INFN-CNAF Bologna, Italy sergio.andreozzi@cnaf.infn.it http://glueschema.forge.cnaf.infn.it 30-31 October 2006 - London - UK

  2. OUTLINE • GLUE Schema 2 • Modeling guidelines • Relationship with JSDL • Relationship with CIM 30-31 October 2006 - London - UK

  3. Modeling Guidelines 30-31 October 2006 - London - UK

  4. Generalization • Generalization based on resource type • e.g.: the query to search for storage space should not depend on the storage system type • The model should accomodate different storage systems, from simple disk-servers to complex SRM-managed systems 30-31 October 2006 - London - UK

  5. Contract-based Design • Resource provision is based on agreements between providers and consumers • The same contract can be implemented using different implementation choices • The different implementation choices should not be exposed 30-31 October 2006 - London - UK

  6. Contract-based design - example • A cluster with 100 CPUs provides 50% of share to VO A and 50% of share to VO B • Two example configurations: • Config 1: • one batch queue with two different shares assigned to two different group ID; • Map each VO user to the related group ID • Config 2: • two batch queues, the share is associated to each queue • Each queue is authorized to one VO • In GLUE Schema 1.2, the two scenarios lead to two different representations in the information service • 1 CE + 2 VOView vs. 2 CE 30-31 October 2006 - London - UK

  7. Contract-based design - example • The same service is provided using two different configuration strategies • Why two different representations at the Grid level? • Contract-based view instead of implementation-based view 30-31 October 2006 - London - UK

  8. Contract-based design - comments • Refine CE concept concentring on: • Isolation of physical level from Grid level • Identification of the core concepts for the Grid level • What is the shared resource? • Execution environment • How the access is policied? • Service differentiation Grid Lv. Phys. Lv. Grid Lv. Phys. Lv. 30-31 October 2006 - London - UK

  9. Denormalization/Redundancy • Information service is read-only from the consumer side • Redundancy can easy the query process • Avoid many-to-many relationship as much as possible possible • Always evaluate the impact of modeling choices on the query process 30-31 October 2006 - London - UK

  10. Heterogeneous Clusters • Since the beginning, the GLUE Schema enables to model subsets of homogenous nodes in a cluster • This feature has been never used because from the Grid level it is not possible to submit a job providing requirements on the worker node characteristics (at least with GT2 GRAM) • This implied the usage of only one subcluster instance to be related to the less powerful machines!!! • How can we better deal with this situation? • One single description with min/avg/max values? • A default description + the different descriptions? • Waiting for more powerful job submission protocols? 30-31 October 2006 - London - UK

  11. Multiple Endpoints to the Same Service • The current schema does not support multiple network endpoints to the same service instance Or • The current schema does not support the representation of a single resource with multiple/alternative service instances to access it • E.g.: a single cluster with two head-nodes/gatekeepers appears as two different clusters? 30-31 October 2006 - London - UK

  12. Multiple Endpoints to the Same Service • Idea: the core concept to be modeled is the resource, not the service • The resource has a URN (Uniform Resource Name) • The resource may have one or more services • A default service/endpoint should be identifiable • If I have multiple service instances to the same resource, how do I choose among them? • Can they be accessible from different VO/groups? 30-31 October 2006 - London - UK

  13. JSDL and GLUE Schema • JSDL: an OGF standard defining an extensible language to describe job submission • Job submission implies expressing requiremetns on resources gLite.JDL : GLUE Schema = OGF.JSDL : RRL To my knowledge, no external RRL for JSDL exists today; an internal set of resource-related attributes exist 30-31 October 2006 - London - UK

  14. Workflow Job Job Job Job JSDL JLM … JSDL JSDL JSDL JLM JLM JLM … … … RRL JPL RRL RRL RRL JPL JPL JPL WS-A SDL … WS-A WS-A WS-A SDL SDL SDL … … … JSDL and GLUE Schema GLUE Schema 2 as RRL for JSDL? 30-31 October 2006 - London - UK RRL - Resource Requirements Language SDL – Scheduling Description Language WS-A – WS-Agreement JLM – Job Lifetime Management JPL – Job Policy Language

  15. JSDL and GLUE Schema • Aspects to be considered: • GLUE Schema is a description from the resource viewpoint • E.g.: Glue.CE.Policy.MaxCPUTime • In gLite.JDL, requirements are expressed as predicate over resource properties • E.g.: GlueCEPolicyMaxCPUTime>600 • In JSDL, the requirements are stated not as predicate, but as attributes related to acceptable thresholds: • <jsdl:TotalCPUTime> • <jsdl:UpperBoundedRange>600.0</jsdl:UpperBoundedRange> • </jsdl:TotalCPUTime> IMPORTANT ISSUE TO BE CONSIDERED 30-31 October 2006 - London - UK

  16. GLUE Schema and CIM /1 • GLUE Schema • tailored approach to the modeling of Grid resources for brokering, access and monitoring • Principles:aggregation, generalization and denormalization • Description in UML, mapping into different data models • HEP-Grid community standard (expanding adoption) • CIM • modeling of IT resources for their management • Principles: fine-grained description and normalization • description in UML, mapping into concrete data models (mainly XML) • DMTF standard • Part of WBEM to provide infrastructure for management comprising interface, protocol and query language to interact with a managed resource 30-31 October 2006 - London - UK

  17. A Dream Solution? • GLUE Schema 2 • Start from 1.3 • Remove deprecated attributes • Clean all the bad design • Address new requirements from new projects • Cast it into the CIM meta-model • Derive a RRL for JSDL 1.x • Derive a schema for the various information services • Write a suite of test queries for the different mappings Is this feasable? 30-31 October 2006 - London - UK

  18. CONCLUSION • We have strong experience on production Grids • The integration to OGF standards is an important goal, nevertheless it is not going to be easy • What is the best way to approach this major evolution? 30-31 October 2006 - London - UK

More Related