Describing and Discovering Mathematical Web and Grid Services Mike - PowerPoint PPT Presentation

The Numerical Algorithms Group Combining mathematics and technology for enhanced performance Describing and Discovering Mathematical Web and Grid Services Mike Dewar NAG Ltd Mike.Dewar@nag.co.uk Over 30 years of mathematical excellence

NAG Background � Develops and sells software for solving complex mathematical problems � libraries of numerical algorithms (Fortran, C) � visualisation environment (IRIS Explorer) � statistical packages (Genstat, GLIM) � compilers and tools (Fortran 95) � Undertakes specialised software engineering tasks � maths libraries for Intel and AMD � Customers in many application areas across industry and academia � Growing business as a component supplier to other software companies Over 30 years of mathematical excellence

NAG and Grid Services � Web/Grid service architecture � Take advantage of new application development paradigm � Service/Software descriptions (MONET) � Semantic web � Brokering grid/web services � Focus on capabilities and properties of software rather than its interface � Portable grid applications for HPC (SciParc) � Platform-independent software layer for grid computing � Take advantage of properties of a given architecture including availability of particular resources and components � Move beyond data grids to algorithm grids � Collaborative Visualisation (gViz Project) Over 30 years of mathematical excellence

MONET – Mathematics On the NET � 2 Year FP5 RTD activity � Principal goals: � Design an architecture for brokering mathematical web services � Develop the necessary mechanisms for describing services, problems and queries � Build prototype brokers and services based on generic technologies � Partners: � NAG Ltd, Stilo Technology Ltd, Universities of Bath, Eindhoven, Manchester and Western Ontario, I3S Nice Over 30 years of mathematical excellence

Simplified MONET Architecture Service Service Service Planning Manager Manager Client BROKER Client Manager Over 30 years of mathematical excellence

MONET Protocols Mathematical Problem consists of � Concrete problem 1. or Problem description 2. Mathematical Query consists of � Problem 1. Logistical information 2. Mathematical Service Description consists of: � Functional Description 1. Implementation Description 2. Service Interface Document 3. Service Binding Document 4. Service Metadata 5. Over 30 years of mathematical excellence

MSDL: Functional Description of Service Reference to Mathematical � Taxonomies � Problem Description Library � e.g. GAMS G2h1a2 Input: � F(v 1 …v n ): R n → R � Provide a hook into UDDI 1. A ⊆ R n 2. � Semantics Supported {D i : R n → R , i=1..n } 3. � OpenMath CDs Output: � x ∈ A 1. � Directives f ∈ R 2. � Find Pre-Conditions � D i = ∂ F/ ∂ v i � Semantic Description 1. Post-Conditions: � � RDFS/OWL + OpenMath F(x) = f 1. There is no y ∈ A | F(y) < f 2. Over 30 years of mathematical excellence

The Matching Process � Query and service descriptions are converted to OWL documents using MONET ontology � Use generic Instance Store component to find service documents which are consistent with the query � Assemble plans for solving problem using generic orchestration language e.g. BPEL4WS � Return those plans to user with details of how closely they match Over 30 years of mathematical excellence

SciParc � Applications: Quantum Chromodynamics, Biophysics, Hydrodynamics, … � Software: well-defined components and building blocks for assembly and deployment of application codes in a grid setting � Flexible and general enough for real applications � Capable of efficient implementation on range of hardware � Modular, extensible structure rather than rigid API � Hardware: massively parallel systems (apeNEXT, BlueGene/L, large clusters, …) Over 30 years of mathematical excellence

Summary � Very little real demand for commercial grid/web service software in our customer base (except .NET!) � Existing tools and infrastructure not really industrial strength � WSDL support limited � OWL tools incomplete � Little in the way of existing ontologies and metadata � However indirect use of MONET-style technology in-house very promising � specification of algorithms/software � used to embed code in different environments/customise codes for particular clients � used to customise documentation � can envisage automated assembly of grid applications based on these specifications Over 30 years of mathematical excellence

Further Information � NAG: � http://www.nag.co.uk � MONET: � http://monet.nag.co.uk � SciParc: � http://www-zeuthen.desy.de/sciparc Over 30 years of mathematical excellence