overview

Grid programming with components: an advanced COMPonent platform for an effective invisible grid

© 2006 GridCOMP Grids Programming with components. An advanced component platform for an effective invisible grid is a Specific Targeted Research Project supported by the IST programme of the European Commission (DG Information Society and Media, project n°034442)

Methodology for Component-based Development of Grid Applications

Artie Basukoski

T.Weigold, P.Buhler – IBM Zurich Research LabA.Basukoski, V.Getov, S.Isiaiadis – University of Westminster

[email protected]

Grid programming with components: an advanced COMPonent platform for an effective invisible grid 2

Overview

GCM and GridCOMP

Component framework overview

The Grid integrated development environment (GIDE)

Biometric identification case study

Initial Experiences


GCM and GridCOMP

GCM

A new advanced Grid Component Model (GCM) providing high level of abstraction and specifically designed for large scale dynamic Grid infrastructures.

Specified within the CoreGRID Euopean project.

GridCOMP

EU project: Grid programming with COMPonents. Design and implementation of a Grid component

framework based on GCM. Includes the development of a Grid IDE and several

use case applications. GCM reference implementation based on ProActive.


GridCOMP Component Framework Overview

The GridCOMP framework is based on the Fractal component model.The main technical features of the component framework are:

Support for primitive and composite distributed components and

hierarchical composition. XML based architecture description language (ADL). Collective interfaces to comply with Grid specific multi-way

communication requirements. A comprehensive runtime API. Support for non-functional aspects such as component control,

skeletons, and autonomicity. Advanced component deployment via the notion of virtual nodes

and deployment descriptors.



Hierarchical composition: all three components can be distributed

controller

content

serverinterface client

interfaces

composite component

primitivecomponent

primitivecomponent

non-functional (control)interfaces

binding



Collective interfaces: The framework takes care of parallel invocations, data distribution, and synchronization.

multicast interface(one-to-one mode)

invocation parameter

scattered parameter invocation parameter

gathercast interface

list ofaggregated parameters

Methodology

A generic method to develop applications based on a Grid Component Model (GCM).

Integrated IDE for Development, Deployment and Steering.

Advantages of our methodology reduced development cycle, increased portability, and support of dynamic properties of the GCM-based

component framework.


Component-Centric Problem-to-Solution Pipeline

Main issues: composition and dynamic properties – deployment, monitoring and steering

Component-based Grid platform design methodology

Monitor andSteer

DeploymentCompositionProgrammingModel - GCM

Applications(Algorithms)

Grid Integrated Development Environment

Metadata Description incl. ADL, etc.

Obtaining theSolution


Strategy: Eclipse Framework for GIDE

Simplify complexity through graphical composition/tools

But, allowing ONLY graphical composition can be inflexible and inefficient

Support for 3 levels of Development Graphical Composition Based on GCM and ProActive (using ADL) Java

Seamless integration with Eclipse Widely supported with many potential plug-ins

“Lets not restrict developers”


Grid IDE Architecture - Core Block Diagram

APPLICATIONS (USE CASES) WP5

IC2D

APIUSES (Requirements)

1.3.1-TestTool

1.3.2-Debug Tool

1.3.3-FinalizationDeployment

Tool

1.2.1-ADL Parser/Verifier

1.2.2-ADL Renderer

1.2.3- Code/ADL Generator

EclipseFramework

Grid IDEToolset

2.1-Monitoring

2.1.1-Component

Monitor

2.1.2-NodeResourceMonitor

2.2-Steering1.2-Composition 1.3-Deployment

1-Development IDE 2-Data Centre IDE

2.2.1- Start/Stop

2.2.2-Install/Remove


GIDE – An Insight into Composition

GIDE builds on GMF for providing graphical front-end The IDE includes

Built-in ADL parser and exporter Verifier Component Repository (allow packaging and sharing) Java source code generation

ADL files are verified, parsed and then appropriate internal representations of compositions (semantic representation) and diagrams are generated.

GIDE delegates the user-interactions to these internal representations


GIDE – An Insight into Composition

Interactive Composition

Composition

Diagram

Importer

Ge

ne

ra

tor

ADL

ADL

ADL

ADL

ADL

ADL

ADL

ADLParser

Parse Tree


Case Study: Biometric Identification System (BIS)

Identify people solely on their biometric information (1:N match)

Use fingerprint biometrics

Consider multiple fingers per person to work reliably on large user population

Use distributed matching to achieve real-time performance

Based on business process (workflow) engine for adaptability



Grid component architecture, bindings, and deployment

1. Biometric matching component CompIDMatcher is deployed on each node

2. DB of known identities is distributed across the nodes

3. Identification requests are broadcasted via multicast interface I2

4. Each node searches its part of the DB

Development – BIS Case Study




<definition name="com.ibm.bis.CompIDMatcher"> <interface name="server" role="server" signature="com.ibm.bis.I3"/> <interface name="client" role="client" signature="com.ibm.bis.I3"/> <component name="CompAlgControl“ definition="com.ibm.bis.CompAlgControl"/> <component name="CompAlg" definition="com.ibm.bis.CompAlg"/> <binding client="this.client" server="CompAlgControl.idServer"/> <binding client="CompAlgControl.client" server="CompAlg.server"/> <binding client="CompAlg.client“ server="CompAlgControl.controlServer"/> <controller desc="composite"/> <virtual-node name="BIS-Grid" cardinality="single"/></definition>

ADL Example: the composite component

CompIDMatcher


Initial Experiences The high level of abstraction provided by the GIDE framework

hides the complexity of Grid programming

The strict separation of concerns (ADL definitions, interface definitions, virtual nodes, deployment descriptors) separates the design from the deployment (physical) infrastructure

The hierarchical nature allows fast composition and facilitates reuse

Drawback: A large number of development artefacts need to be generated and maintained. This makes manual code re-factoring complex.

Need to gather development statistics in order to accurately gauge the impact of the GIDE development methodology on the development lifecycle.

The GIDE, implemented as an Eclipse plug-in. Project’s WIKI: http://perun.hscs.wmin.ac.uk/dis/gide/

overview

Documents

grid component framework

advanced component platform

grid ide

advanced component deployment

component control

gcmbased component framework

fractal component model

development cycle