Towards GLUE 2: Evolution of the Computing Element ... - NorduGrid

International Conference on Computing in High Energy and Nuclear Physics (CHEP’07) IOP Publishing Journal of Physics: Conference Series 119 (2008) 062009 doi:10.1088/1742-6596/119/6/062009

Towards GLUE 2: Evolution of the Computing Element Information Model S Andreozzi1 , S Burke2 , L Field3 and B K´ onya4 1 2 3 4

INFN-CNAF, Viale Berti Pichat 6/2, 40126 Bologna, Italy Rutherford Appleton Laboratory, Didcot, OX11 0QX, UK CERN, CH-1211 Genve 23, Switzerland Lund University, Box 117, S-221 00 Lund, Sweden

E-mail: [email protected], [email protected], [email protected], [email protected] Abstract. A key advantage of Grid systems is the ability to share heterogeneous resources and services between traditional administrative and organizational domains. This ability enables virtual pools of resources to be created and assigned to groups of users. Resource awareness, the capability of users or user agents to have knowledge about the existence and state of resources, is required in order utilize the resource. This awareness requires a description of the services and resources typically defined via a community-agreed information model. One of the most popular information models, used by a number of Grid infrastructures, is the GLUE Schema, which provides a common language for describing Grid resources. Other approaches exist, however they follow different modeling strategies. The presence of different flavors of information models for Grid resources is a barrier for enabling inter-Grid interoperability. In order to solve this problem, the GLUE Working Group in the context of the Open Grid Forum was started. The purpose of the group is to oversee a major redesign of the GLUE Schema which should consider the successful modeling choices and flaws that have emerged from practical experience and modeling choices from other initiatives. In this paper, we present the status of the new model for describing computing resources as the first output from the working group with the aim of dissemination and soliciting feedback from the community.

1. Introduction Grid systems enable the sharing of distributed resources across multiple heterogeneous platforms, locations, and organizations. A precise and shared description of these resources needs to be defined in order to discover their characteristics and status. This description needs to be common for different Grid middleware and infrastructures in order to facilitate interoperability. A description of Grid resources is typically performed in two steps. Firstly, the modeling activity concentrates on capturing the abstract representation of entities in an implementationindependent fashion. This step is supported by languages such as Unified Modeling Language (UML) and the result is called an Information Model. The main types of modeling concepts are entities, their properties, operations and relationships. The information model can be then rendered into one or many different data models which represents the information model in a given language and/or a specification. It is the data models which enables the information model to transmitted, manipulated and accessed [1].

c 2008 IOP Publishing Ltd 

1

International Conference on Computing in High Energy and Nuclear Physics (CHEP’07) IOP Publishing Journal of Physics: Conference Series 119 (2008) 062009 doi:10.1088/1742-6596/119/6/062009

Figure 1. Timeline A number of information models and data models already exist for Grid resources. The most wide spread solution is the GLUE Schema [2] which was defined as a common description for resources between European and US Grid projects and is currently used by the EGEE [3] and OSG [4] Grid infrastructures. Another solution is the NorduGrid Schema [5]. These two approaches are being unified in the context of the Open Grid Forum as a community standard defined by the OGF GLUE Working Group. This paper presents the current status and the future prospects of this activity. 2. The OGF GLUE Working Group In the context of the Open Grid Forum, the GLUE Working Group [5] was created in January 2007 to provide a recommendation for an abstract information model which leverages the existing proposals into a community-agreed standard. The group plans to also provide reference implementations for the relevant concrete data models required by the community (e.g., XML Schema, relational, LDAP, RDF). Such implementations will consider simplifications specific to the target data model in order to improve query performance or other aspects. The activity benefits from several years of experience in the context of production Grid systems such as such as EGEE, OSG and NorduGrid. Other infrastructure projects are supporting and contributing to this activity, in particular TeraGrid [6], NAREGI [7], NGS [8], D-GRID [9], and APACGrid [10]. Relevant contributions have also been given by middlewareoriented projects such as OMII-Europe [11], KnowArc [12] and UNICORE [13]. Two main documents will be produced [14]: a use case document describing the production scenarios that motivate the need for the descriptions and a specification document with the precise definition of classes, attributes and relationships in response to the given use cases. In the first year of activity, the group will focus on a redesign of models for the following main concepts: administrative domains, services, computing resources and storage resources. The timeline in Figure 1 shows the tentative schedule for the work. 2.1. Relationships to other OGF Working Groups The description of Grid resources is a key piece of information that is relevant to other OGF activities. The BES (Basic Execution Service) implementations can expose a richer description of the computing resources by means of the GLUE computing model. A JSDL (job Submission Description Language) document can exploit the GLUE terms to enable users to express requirements on resources. The SAGA (Simple Access Grid API) service discovery API can enable the searching for services based on the GLUE Service specification. The GSM (Grid Storage Management) can expose a richer description of storage resources managed by an SRM (Storage Resource Manager) using the GLUE Storage model. GLUE contributes to the information model architecture being defined by the OGSA Resource Management Design Team. GLUE should also be coherent with the Reference Model for Grid systems 2.

2

International Conference on Computing in High Energy and Nuclear Physics (CHEP’07) IOP Publishing Journal of Physics: Conference Series 119 (2008) 062009 doi:10.1088/1742-6596/119/6/062009

Figure 2. Relationship to other OGF Working Groups 3. Relationship to CIM CIM (Common Information Model) is a standard defined by the DMTF (Distributed Management Task Force) [15] with the goal of providing a common definition of management information for systems, networks, applications and services, and allows for vendor extensions. Such definitions should enable vendors to exchange semantically rich management information between systems throughout the network. By leveraging the OGF Liaison with DMTF, the GLUE WG is engaging with the DMTF in order to leverage CIM and other DMTF standards for the modeling of GLUE 2. The goal is to render the GLUE 2 specification in terms of the CIM meta-model in order to exploit the WBEM technologies for a possible implementation. 4. GLUE 2 In this section, the main entities defined in the current draft information model for GLUE 2 are described. They are subject to change in the future revision. In Section 4.1, we present the core entities while in Section 4.2 we describe the concepts related to a computing capability. The Storage Element model is also a goal of the OGF GLUE Working Group, nevertheless the description of the current status of the activity is out of scope in this paper. 4.1. The Main Entities The main entities that are central to the GLUE Information model are the concepts of Service, Activity, Resource, Element and AdminDomain (see Figure 3). A Service is defined as an abstracted, logical view of actual software components having a well-defined interface and offering one or more functionalities. A Service typically exposes one or more network endpoints. An Activity is a unit of work managed by a service. An activity can have relationships to other activities being managed by different services, therefore it shares a common context. A Resource is an entity that is useful in a Grid environment and that is offered through one or more service). A resource usually represents aggregated information. The Element enables to group concepts that participate in the creation of an entity useful in a Grid environment. The element is autonomous and can be composed by services and the exposed resources. The simplest element is the Service Element, that is composed by a single service. The AdminDomain is used to identifies atomic management units responsible for a set of elements. Elements part of an administrative domain can span different physical locations. 3

International Conference on Computing in High Energy and Nuclear Physics (CHEP’07) IOP Publishing Journal of Physics: Conference Series 119 (2008) 062009 doi:10.1088/1742-6596/119/6/062009

Figure 3. Main Entities 4.2. The Computing Information Model An important type of element useful in a Grid environment is the Computing Element (see Figure 4). A Computing Element groups the concepts that participate in the creation of an entity providing computational activity in a Grid environment. This entity is atomic in the sense that it cannot be decomposed in smaller computing elements from the Grid viewpoint. The following entities have been identified as useful to be modeled for describing the structure of a computing element: computing service, computing resource, computing share, execution environment, application environment and job. The Computing Service is a specialization of a service for creating, monitoring and controlling computational activities called jobs. The Execution Environment provides a description of hardware and software characteristics that define a type of environment available to and requestable by a Grid job when submitted to a Computing Service. Such a description also includes information about the total/available/used instances of the execution environment. An Execution Environment may also contain one or more Application Environments. The Computing Resource is a grouping concept for a set of different types of execution environments; the aggregation is defined by the common management scope (e.g., a local resource management system like a batch system defines an aggregation scope). An important concept for a Computing Element is the Computing Share, which is a utilization target for a set of computing resources defined by policies (Computing Share Policy) and characterized by status information (Computing Share State). 5. Implementations Several implementations of the GLUE 2 information model are expected. The OMII-Europe project is an important contributor in this phase as it is contributing not only to the definition of the information model, but also an implementation for the computing element to be integrated into the OGSA-BES implementations of gLite, UNICORE and Globus [16]. Such an implementation is based on WBEM technologies and aims to add a management layer. It is hoped that an alpha release will be demonstrated at SuperComputing 2007 [17], while the final implementation will be available by the end of the project (April 2008). The KnowArc project will also work on implementing and integrating GLUE 2 into the next generation Advanced Resource Connector (ARC) middleware.

4

International Conference on Computing in High Energy and Nuclear Physics (CHEP’07) IOP Publishing Journal of Physics: Conference Series 119 (2008) 062009 doi:10.1088/1742-6596/119/6/062009

Figure 4. Computing Element 6. Conclusion The unification of existing description of Grid resources into an open standard is an essential building block for interoperable Grid infrastructures. In this paper, we have presented the motivation and the activity of the OGF GLUE Working Group. The current view of the main concepts and for the computing element were presented together with references to on-going implementation activity. 7. Acknowledgements We thank all the persons that are contributing to the OGF GLUE WG as their contribution is essential to the success of this work. We also thank Ellen Stokes, Hiro Kishimoto and Tom Roney for their guidance in the OGF community and for the support with external relationships with DMTF. The final thank is for Paul Strong from the Reference Model WG for the fruitful discussion about integrating GLUE and the Reference Model. References [1] Maciel F 2007 Guidelines for Information Modeling for OGSA Entities OGF Draft Document [2] Andreozzi S, Burke S and Field L 2007 Proceedings of the International Conference on Computing in High Energy and Nuclear Physics (CHEP2007), Victoria, Canada [3] EGEE: European Grid for E-sciencE http://www.eu-egee.org [4] OSG: Open Science Grid http://www.osg.org [5] OGF GLUE Working Group http://forge.ogf.org/sf/sfmain/do/viewProject/projects.glue-wg [6] TERAGRID http://www.teragrid.org/ [7] NAREGI http://www.naregi.org/download/ [8] NGS - UK National Grid Service http://www.grid-support.ac.uk/ [9] D-Grid: Die Deutsche Grid-Initiative http://www.d-grid.de/ [10] APACGrid http://grid.apac.edu.au/ [11] The OMII-Europe project http://omii-europe.org [12] The KnowArc project http://www.knowarc.org/ [13] UNICORE: Uniform Interface to Computing Resources) http://www.unicore.eu/ [14] GLUE Use Cases and Specification document drafts http://forge.ogf.org/sf/docman/do/listDocuments/projects.gluewg/docman.root.drafts

5

International Conference on Computing in High Energy and Nuclear Physics (CHEP’07) IOP Publishing Journal of Physics: Conference Series 119 (2008) 062009 doi:10.1088/1742-6596/119/6/062009 [15] Distributed Management Task Force http://www.dmtf.org [16] GLUE 2 and OGSA-BES Integration. http://omii-europe.forge.cnaf.infn.it/jra2/glue/bes-glue-integration [17] SuperComputing 2007 Conference http://sc07.supercomputing.org/

6