75
The Grid - Multi- The Grid - Multi- Domain Domain Distributed Distributed Computing Computing Kai Rasmussen Kai Rasmussen Paul Ruggieri Paul Ruggieri

The Grid - Multi-Domain Distributed Computing Kai Rasmussen Paul Ruggieri

Embed Size (px)

Citation preview

The Grid - Multi-DomainThe Grid - Multi-DomainDistributed ComputingDistributed Computing

Kai RasmussenKai Rasmussen

Paul RuggieriPaul Ruggieri

Topic OverviewTopic Overview

The GridThe Grid TypesTypes Virtual OrganizationsVirtual Organizations SecuritySecurity Real ExamplesReal Examples

Grid ToolsGrid Tools CondorCondor CactusCactus Cactus-GCactus-G GlobusGlobus OGSAOGSA

The GridThe Grid

What is a Grid system?What is a Grid system? Highly heterogeneous set of resources that Highly heterogeneous set of resources that

may or may not be maintained by multiple may or may not be maintained by multiple administrative domainsadministrative domains

Early ideaEarly idea Computational resources would be universally Computational resources would be universally

available as electric poweravailable as electric power

““A hardware and software infrastructures that provides A hardware and software infrastructures that provides dependable, consistent, pervasive and inexpensive dependable, consistent, pervasive and inexpensive access to high-end computational capabilities” - Ian access to high-end computational capabilities” - Ian FosterFoster

• Resources are distributed across sites and Resources are distributed across sites and organizations with no centralized point of controlorganizations with no centralized point of control

• What constitutes a Grid?What constitutes a Grid?• Resources coordinated without being subjected to a Resources coordinated without being subjected to a

centralized controlcentralized control• Uses standard, open source protocols and interfacesUses standard, open source protocols and interfaces• Delivers non-trivial qualities of serviceDelivers non-trivial qualities of service

Grid TypesGrid Types

Computations Grids Computations Grids Resource pure CPUResource pure CPU Strength: Computational Intensive applicationsStrength: Computational Intensive applications

Data GridsData Grids Shared storage and dataShared storage and data Terabytes of storage space.Terabytes of storage space. Sharing of data among collaboratorsSharing of data among collaborators Fault ToleranceFault Tolerance

Equipment GridsEquipment Grids Set of resources that surround shared equipments, Set of resources that surround shared equipments,

such as a telescopesuch as a telescope

Virtual OrganizationsVirtual Organizations

Grids are Multi-domainGrids are Multi-domain Resources are administrated by separate Resources are administrated by separate

departments or institutionsdepartments or institutions All wish to maintain individual controlAll wish to maintain individual control

There is a cross site grouping of collaborators There is a cross site grouping of collaborators sharing resourcessharing resources ““Virtual Organization”Virtual Organization”

Virtual OrganizationsVirtual Organizations

Users of VO’s share a common goal and Users of VO’s share a common goal and trusttrust

Collection of resources, users and rules Collection of resources, users and rules governing sharinggoverning sharing Highly controlled - What is Shared? Who is Highly controlled - What is Shared? Who is

Sharing? How can resources be used?Sharing? How can resources be used?

One global domains acting over individual One global domains acting over individual collaborating domainscollaborating domains

Grid SecurityGrid Security

Highly distributed natureHighly distributed nature VOs spread over many security domainsVOs spread over many security domains

AuthenticationAuthentication Proving identityProving identity

AuthorizationAuthorization Obtaining privilegesObtaining privileges

Confidentiality & IntegrityConfidentiality & Integrity Identity and privileges can be trustedIdentity and privileges can be trusted

AuthenticationAuthentication

Certificate Authority (CA)Certificate Authority (CA) Entity that signs certificate that proves users identityEntity that signs certificate that proves users identity Certificate then used as credentials to use system Certificate then used as credentials to use system Typically several CAs to prevent single point of Typically several CAs to prevent single point of

failure/attackfailure/attack Globus Grid Security Infrastructure (GSI)Globus Grid Security Infrastructure (GSI)

Globus’s Authentication componentGlobus’s Authentication component Global security credential later mapped to localGlobal security credential later mapped to local

Kerberos tickets or local username and passwordKerberos tickets or local username and password Typically generate short-term proxy certificate with Typically generate short-term proxy certificate with

long-term certificatelong-term certificate

AuthenticationAuthentication

Certification Authority Coordination GroupCertification Authority Coordination Group Maintains a global infrastructure of trusted CA agentsMaintains a global infrastructure of trusted CA agents CA must meet standardsCA must meet standards

Physically securePhysically secure Must validate identity with Registration Authorities using Must validate identity with Registration Authorities using

official documents or photographic identificationofficial documents or photographic identification Private Keys must be minimum of 1020 Bits and have max 1 Private Keys must be minimum of 1020 Bits and have max 1

year lifeyear life

28 approved CAs is European union28 approved CAs is European union

Security IssuesSecurity Issues

DelegationDelegation User entrusts separate entity to perform taskUser entrusts separate entity to perform task

Entity must be given certification and trusted to Entity must be given certification and trusted to behavebehave

Limit proxies strengthLimit proxies strength Endow proxy with specific purposeEndow proxy with specific purpose

Grid ProjectsGrid Projects

EGEE - Enabling Grids for eScienceEGEE - Enabling Grids for eScience 70 sites in over 27 countries70 sites in over 27 countries

Mostly EuropeanMostly European

40 Virtual Organizations40 Virtual Organizations GENIUS Grid-Portal is used for submissionGENIUS Grid-Portal is used for submission

Individual collaborators use own middle-ware tools Individual collaborators use own middle-ware tools to group resourcesto group resources

LCGLCG

Large Hadron Collider Computation GridLarge Hadron Collider Computation Grid Developed distributed systems needed to Developed distributed systems needed to

support computation and data needs of LHC support computation and data needs of LHC physics experimentsphysics experiments

EGEE CollaboratorEGEE Collaborator 100 Sites100 Sites Worlds largest GridWorlds largest Grid

Grid 2003Grid 2003

US effortUS effort 27 National sites27 National sites 28000 Processors, 13000 Simultaneous Jobs28000 Processors, 13000 Simultaneous Jobs Infrastructure forInfrastructure for

Particle Physics GridParticle Physics Grid Virtual Data Grid LaboratoryVirtual Data Grid Laboratory

Develop Application Grid Laboratory - Grid3Develop Application Grid Laboratory - Grid3 Platform for experimental CS ResearchPlatform for experimental CS Research

Built on Virtual Data ToolkitBuilt on Virtual Data Toolkit Collection of Globus, Condor and other middleware toolsCollection of Globus, Condor and other middleware tools

TeraGridTeraGrid

40 Teraflops of Computational Power40 Teraflops of Computational Power 8 National Sites with strong backbone8 National Sites with strong backbone Used for NSF sponsored High Used for NSF sponsored High

Performance ComputingPerformance Computing Mapping the human arterial tree modelMapping the human arterial tree model TeraShake - Earthquake simulationTeraShake - Earthquake simulation

ApplicationsApplications

Climate Monitoring + SimulationClimate Monitoring + Simulation Network Weather ServiceNetwork Weather Service Climate Data-Analysis ToolClimate Data-Analysis Tool

Both run on the Earth System Grid running on GlobusBoth run on the Earth System Grid running on Globus

MEANDER nowcast meteorologyMEANDER nowcast meteorology Run on Hungarian SupergridRun on Hungarian Supergrid

ATLAS ChallengeATLAS Challenge Simulate high energy proton-proton collisionsSimulate high energy proton-proton collisions

Computational Science SimulationsComputational Science Simulations Biology, Fluid DynamicsBiology, Fluid Dynamics

Grid ToolsGrid Tools

Many middleware implementationsMany middleware implementations GlobusGlobus CondorCondor

Condor-GCondor-G Cactus-GCactus-G OGSAOGSA

Solves common Grid problemsSolves common Grid problems Resource discovery/management/allocationResource discovery/management/allocation Security/AuthenticationSecurity/Authentication

CondorCondor

Initially developed in 1983 at University of Initially developed in 1983 at University of WisconsinWisconsin Pre-Grid toolPre-Grid tool A Local Resource Management SystemA Local Resource Management System

Allows creation of communities with distributed Allows creation of communities with distributed resourcesresources Communities should grown naturallyCommunities should grown naturally Sharing as much or as little as they care tooSharing as much or as little as they care too Sounds like Virtual OrganizationsSounds like Virtual Organizations

CondorCondor

ResponsibilitiesResponsibilities Job Management, SchedulingJob Management, Scheduling Resource monitoring and managementResource monitoring and management Checkpointing and MigrationCheckpointing and Migration Utilize idle CPUUtilize idle CPU

Cycle ‘ScavengeCycle ‘Scavenge

Condor PoolCondor Pool

Full set of users and resources in Full set of users and resources in communitycommunity Composed of three EntitiesComposed of three Entities

AgentAgent Finds resources and executes jobsFinds resources and executes jobs

ResourceResource Advertise itself and how it can be used in poolAdvertise itself and how it can be used in pool

MatchmakerMatchmaker Knows of all agents and resourcesKnows of all agents and resources Puts together compatible pairsPuts together compatible pairs Pool is defined by single matchmakerPool is defined by single matchmaker

MatchmakingMatchmaking

Problem of centralized SchedulingProblem of centralized Scheduling Resources have multiple ownersResources have multiple owners Unique use requirementsUnique use requirements

Matchmaking finds balance between user Matchmaking finds balance between user and resource needsand resource needs

ClassAdsClassAds Agents advertise requirementsAgents advertise requirements Resources advertise how it can be usedResources advertise how it can be used

MatchmakingMatchmaking

Matchmaker scans all known ClassAdsMatchmaker scans all known ClassAds Creates matching pairs of agents and resourcesCreates matching pairs of agents and resources Informs both partiesInforms both parties

Individually responsible to negotiate job and initiating Individually responsible to negotiate job and initiating execution of jobexecution of job

Separation of matching and claimingSeparation of matching and claiming Matchmaker unaware of complicated allocationMatchmaker unaware of complicated allocation Stale information may exist. Resource can deny Stale information may exist. Resource can deny

matchmatch

Condor FlockingCondor Flocking

Linking condor pools necessary for Linking condor pools necessary for collaborationcollaboration Sharing of resources beyond the Sharing of resources beyond the

organizational levelorganizational level Individuals belonging to multiple communitiesIndividuals belonging to multiple communities

Gateway FlockingGateway Flocking Entire communities are linkedEntire communities are linked

Direct FlockingDirect Flocking Individual collaborators belong to many poolsIndividual collaborators belong to many pools

Gateway FlockingGateway Flocking

Gateway entity serves as a singular point Gateway entity serves as a singular point of access for cross pool communicationof access for cross pool communication Matchmakers talk to GatewaysMatchmakers talk to Gateways Gateways talk to GatewaysGateways talk to Gateways

Transparent to userTransparent to user Organizational level sharingOrganizational level sharing Powerful, but difficult to setup and Powerful, but difficult to setup and

maintainmaintain

Gateway FlockingGateway Flocking

Direct FlockingDirect Flocking

Agents report to multiple matchmakersAgents report to multiple matchmakers Individual collaborationIndividual collaboration Natural idea for usersNatural idea for users Less powerful but simpler to build and deployLess powerful but simpler to build and deploy

Eventually used in favor Gateway FlockingEventually used in favor Gateway Flocking

Direct FlockingDirect Flocking

CactusCactus

General-purpose, open-source parallel General-purpose, open-source parallel computation frameworkcomputation framework Developed for numerical solution to Einstein’s Developed for numerical solution to Einstein’s

equationequation Two main components flesh and thornsTwo main components flesh and thorns

Flesh – central coreFlesh – central core Thorns – application modulesThorns – application modules

Provides simple abstract APIProvides simple abstract API Hides MPI parallel driver, I/O (thorns)Hides MPI parallel driver, I/O (thorns)

Cactus-GCactus-G

““Grid-enabled” CactusGrid-enabled” Cactus Combines Cactus and MPICH-G2 Combines Cactus and MPICH-G2 (more later)(more later)

Layered approachLayered approach Application thornsApplication thorns Grid-aware infrastructure thornsGrid-aware infrastructure thorns Grid-enabled communication library Grid-enabled communication library (MPICH-G2 in this (MPICH-G2 in this

case)case)

GlobusGlobus

CondorCondor Pre-Grid tool applied to Grid SystemsPre-Grid tool applied to Grid Systems Multi-domain possible but limitedMulti-domain possible but limited No security. Focus primarily on resource No security. Focus primarily on resource

managementmanagement GlobusGlobus

Set of Grid specific toolsSet of Grid specific tools Extendable and HierarchicalExtendable and Hierarchical

The ToolkitThe Toolkit

Globus ToolkitGlobus Toolkit Components for basic security, resource Components for basic security, resource

management, etcmanagement, etc Well defined interfaces - “Hour-glass” Well defined interfaces - “Hour-glass”

architecturearchitecture Local services sit behind APILocal services sit behind API Global services built on top of these local servicesGlobal services built on top of these local services

Interfaces useful to manage heterogeneityInterfaces useful to manage heterogeneity Information Service integral componentInformation Service integral component

Information-rich environment neededInformation-rich environment needed

Globus ServicesGlobus Services

Resource ManagementResource Management

Globus Resource Allocation Manager Globus Resource Allocation Manager (GRAM)(GRAM) Responsible for set of local resourcesResponsible for set of local resources Single domainSingle domain Implemented with set a local RM toolsImplemented with set a local RM tools

Condor, NQE, Fork, Easy-LL, etc…Condor, NQE, Fork, Easy-LL, etc…

Resource requests expressed in Resource requests expressed in Resource Specification Language (RSLResource Specification Language (RSL

Resource BrokerResource Broker

Manages RSL requestsManages RSL requests Uses Information services to discover Uses Information services to discover

GRAMSGRAMS

Transforms abstract RSLs into more Transforms abstract RSLs into more specific requirementsspecific requirements

Sends allocation requests to appropriate Sends allocation requests to appropriate GRAMGRAM

Information ServiceInformation Service

Grid always in fluxGrid always in flux Information rich system produces information users Information rich system produces information users

find usefulfind useful Enhances flexibility and performanceEnhances flexibility and performance

Necessity for administrationNecessity for administration Globus Metacomputing Directory Service (MDS)Globus Metacomputing Directory Service (MDS)

Stores and makes accessible Grid informationStores and makes accessible Grid information Lightweight Directory Access Protocol (LDAP)Lightweight Directory Access Protocol (LDAP)

Extensible representation for informationExtensible representation for information Stores component information in directory information Stores component information in directory information

treetree

SecuritySecurity

Local HeterogeneityLocal Heterogeneity Resources operated in multiple security Resources operated in multiple security

domainsdomains All use different authentication techniquesAll use different authentication techniques

N-Way authenticationN-Way authentication Job may be any number of processes on any Job may be any number of processes on any

number of resourcesnumber of resources One logical entity. User should only One logical entity. User should only

authenticate once.authenticate once.

SecuritySecurity

Globus Security Infrastructure (GSI)Globus Security Infrastructure (GSI) Modular design constructed on top of local Modular design constructed on top of local

servicesservices Solves local heterogeneitySolves local heterogeneity

Globus IdentityGlobus Identity Mapped into local user identities by local GSIMapped into local user identities by local GSI Allows for n-way authorizationAllows for n-way authorization

OGSAOGSA

Open Grid Services ArchitectureOpen Grid Services Architecture Defines a Grid ServiceDefines a Grid Service

Provides standard interface for naming, creating, Provides standard interface for naming, creating, discovering a Grid Servicediscovering a Grid Service Location TransparentLocation Transparent

Globus ToolkitGlobus Toolkit GRAM – resource allocation/managementGRAM – resource allocation/management MDS-2 – information discoveryMDS-2 – information discovery GSI – authentication (single sign-on)GSI – authentication (single sign-on)

Web servicesWeb services Widely usedWidely used Language/system independentLanguage/system independent

OGSA – Grid Service InterfaceOGSA – Grid Service Interface

OGSA – VO StructureOGSA – VO Structure

Condor-GCondor-G

Hybrid Condor-Globus SystemHybrid Condor-Globus System Local Condor agent (Condor-G)Local Condor agent (Condor-G)

Communicates with Globus GRAM, MDS, GSI, etcCommunicates with Globus GRAM, MDS, GSI, etc Optimized Globus’s GRAM to work with Condor Optimized Globus’s GRAM to work with Condor

betterbetter

Specific TestbedSpecific Testbed

Grid2003Grid2003 Organized into 6 VOs (one for each Organized into 6 VOs (one for each

application)application) At each VO site, middleware installed with grid At each VO site, middleware installed with grid

certificate databasescertificate databases GSI, GRAM, and GridFTP used from GlobusGSI, GRAM, and GridFTP used from Globus MDSMDS MonALISA MonALISA

Agent-based monitoring used in conjunction with Agent-based monitoring used in conjunction with MDS MDS

MPICH-G2: A Grid-Enabled MPICH-G2: A Grid-Enabled Implementation of the Implementation of the

Message Passing InterfaceMessage Passing Interface

Nicholas Karonis, Brian Toonen, Ian FosterNicholas Karonis, Brian Toonen, Ian Foster

AbstractAbstract

Grid Enabled MPI implementationGrid Enabled MPI implementation Extends MPICHExtends MPICH Utilizes Globus ToolkitUtilizes Globus Toolkit

Authentication, Authorization, Resource Authentication, Authorization, Resource Allocation, Executable Staging, I/O, Process Allocation, Executable Staging, I/O, Process management creation and controlmanagement creation and control

Hide/Expose critical aspects of Hide/Expose critical aspects of heterogeneous environmentheterogeneous environment

The ProblemThe Problem

Grids difficult to program for… Grids difficult to program for… heterogeneous, highly distributedheterogeneous, highly distributed

Build on existing MPI APIBuild on existing MPI API MPICH specificallyMPICH specifically

Can we implement MPI constructs in a highly Can we implement MPI constructs in a highly heterogeneous environment efficiently and heterogeneous environment efficiently and transparently?transparently? Yes, use Globus!Yes, use Globus!

Can we also allow users to manage Can we also allow users to manage heterogeneity?heterogeneity? Yes, existing MPI Communicator Construct!Yes, existing MPI Communicator Construct!

MPICH-G2MPICH-G2

Global Security Infrastructure (GSI)Global Security Infrastructure (GSI) Single sign-on authenticationSingle sign-on authentication

Monitoring and Discovery Service (MDS)Monitoring and Discovery Service (MDS) Select nodes to execute onSelect nodes to execute on

Resource Specification LanguageResource Specification Language Generated by mpirunGenerated by mpirun Specifies job resource requirementsSpecifies job resource requirements

Dynamically-Updated Request Online Dynamically-Updated Request Online Coallocator (DUROC)Coallocator (DUROC)

MPICH-G2 Flow DiagramMPICH-G2 Flow Diagram

MPICH-G2 ImprovementsMPICH-G2 Improvements

Replaces MPICH-GReplaces MPICH-G Replace use of Nexus (Globus) for all communication Replace use of Nexus (Globus) for all communication

with optimized codewith optimized code Increased BandwidthIncreased Bandwidth

Cutout extra layer (Nexus)Cutout extra layer (Nexus)

Reduce intra-machine vendor MPI messaging latencyReduce intra-machine vendor MPI messaging latency Eliminate unnecessary polling based on source rank info (for Eliminate unnecessary polling based on source rank info (for

Recv)Recv) Specified, Specified-pending, multimethod (more later)Specified, Specified-pending, multimethod (more later) Only poll TCP (expensive) when necessary (ie using TCP not Only poll TCP (expensive) when necessary (ie using TCP not

vendor MPI)vendor MPI)

MPICH-G2 Improvements 2MPICH-G2 Improvements 2

More efficient use of socketsMore efficient use of sockets Uses one socket for both directionsUses one socket for both directions

Multilevel topology-aware collective Multilevel topology-aware collective operationsoperations Collective operations originally implemented Collective operations originally implemented

assuming equidistanceassuming equidistance Not likely in Grid scenarioNot likely in Grid scenario

App Heterogeneity ManagementApp Heterogeneity Management

Topology DiscoveryTopology Discovery Need method of discovering topology to minimize Need method of discovering topology to minimize

expensive transfers expensive transfers intra-site communication vs intra-machine intra-site communication vs intra-machine

communicationcommunication

Use existing MPI communicator constructUse existing MPI communicator construct Associate Associate attributesattributes with communicators with communicators Topology depths and colorsTopology depths and colors

Allow MPI developers to create communicators which Allow MPI developers to create communicators which group processes topologicallygroup processes topologically

Example MPICH-G2 AppExample MPICH-G2 App

Performance GroupingsPerformance Groupings

SpecifiedSpecified MPI_Recv explicitly specifies process on same machineMPI_Recv explicitly specifies process on same machine No outstanding asynchronous operationsNo outstanding asynchronous operations Explicitly call vendor MPIExplicitly call vendor MPI

Specified-pendingSpecified-pending MPI_Recv explicitly specifies process on same machineMPI_Recv explicitly specifies process on same machine Outstanding recv requests on same machineOutstanding recv requests on same machine Forced to continuously poll vendor MPIForced to continuously poll vendor MPI

MultimethodMultimethod MPI_Recv source rank is MPI_ANY_SOURCEMPI_Recv source rank is MPI_ANY_SOURCE OR outstanding recv requests which may require TCPOR outstanding recv requests which may require TCP Forced to continuously poll vendor MPI and TCPForced to continuously poll vendor MPI and TCP

Vendor MPI ResultsVendor MPI Results

Increased performance compared to Increased performance compared to MPICH-GMPICH-G

Relatively close performance to straight Relatively close performance to straight vendor MPIvendor MPI

Vendor MPI ResultsVendor MPI Results

TCP/IP ResultsTCP/IP Results

Similar results as Vendor MPI Similar results as Vendor MPI (less interesting)(less interesting) Authors explicitly say they did not attempt to modify Authors explicitly say they did not attempt to modify

the TCP codethe TCP code

TCP/IP ResultsTCP/IP Results

ConclusionsConclusions

Good performanceGood performance Improved performance opposed to previous versionImproved performance opposed to previous version ““good enough” performance to justify usegood enough” performance to justify use

Eases transition of MPI applications to the Eases transition of MPI applications to the context of a Gridcontext of a Grid Just worksJust works Provides developer with a relatively simply means of Provides developer with a relatively simply means of

writing “smart” apps which are aware of their topologywriting “smart” apps which are aware of their topology

P-GRADE PortalP-GRADE Portal

MTA SZTAKIMTA SZTAKI

Computer and Automation Research Computer and Automation Research Institute of the Hungarian Academy of Institute of the Hungarian Academy of SciencesSciences Laboratory of Parallel and Distributed Laboratory of Parallel and Distributed

ComputingComputing Peter KacsukPeter Kacsuk Joszef PatvarczkiJoszef Patvarczki

HunGridHunGrid Member of both SEE-Grid and EGEEMember of both SEE-Grid and EGEE

Two Grid ProblemsTwo Grid Problems

Middleware tools build together into a GridMiddleware tools build together into a Grid Too many complex partsToo many complex parts Confusing for users with little experienceConfusing for users with little experience

Mostly research scientistsMostly research scientists

PVM and MPI allow for Parallel executionPVM and MPI allow for Parallel execution Executed within a Globus or Condor site Executed within a Globus or Condor site

shows good performanceshows good performance Performance decreases when executed in Performance decreases when executed in

multiple sitesmultiple sites

P-GRADE PortalP-GRADE Portal

A Web based Portal for accessing GridA Web based Portal for accessing Grid High level tools hide complexity of middlewareHigh level tools hide complexity of middleware Can be accessed anywhereCan be accessed anywhere

Workflow solutionWorkflow solution Complex problems are broken into several parts Complex problems are broken into several parts

treated as single frameworktreated as single framework Executed as an acyclic graphExecuted as an acyclic graph Parallelism at two levelsParallelism at two levels

Independent branches run on several grid sitesIndependent branches run on several grid sites Individual nodes can be parallel programs (MPI or PVM)Individual nodes can be parallel programs (MPI or PVM)

PortalPortal

Fully functional; built upon middleware Fully functional; built upon middleware toolstools Grid Certificate managementGrid Certificate management Setting up Grid environmentSetting up Grid environment Creation and modification of workflow appsCreation and modification of workflow apps Management and parallel execution of Management and parallel execution of

workflow apps on grid resourcesworkflow apps on grid resources Visualization of workflow progressVisualization of workflow progress

Grid CertificateGrid Certificate

Security done through Globus GSISecurity done through Globus GSI Connect to Proxy server; download CertificateConnect to Proxy server; download Certificate Monitor statusMonitor status

Resource ManagementResource Management

Use Globus tools to attach jobs to Use Globus tools to attach jobs to resourcesresources Two StrategiesTwo Strategies

Static AllocationStatic Allocation Connect Directly to GRAM ServersConnect Directly to GRAM Servers

Dynamic AllocationDynamic Allocation Connect to MDS serviceConnect to MDS service Allocate through Grid resource brokerAllocate through Grid resource broker

Workflow Creation & Workflow Creation & MonitoringMonitoring

P-GRADEP-GRADE Java app for Java app for

creating parallel creating parallel workflowsworkflows

Directed input Directed input and output filesand output files

Parameter StudyParameter Study

Singular job run under varying input Singular job run under varying input parametersparameters Outputs later compared against each otherOutputs later compared against each other

Logical Grid ApplicationLogical Grid Application Each job independent and can be run in Each job independent and can be run in

parallelparallel

P-GRADE Portal w/ PStudyP-GRADE Portal w/ PStudy

Adaped Portal to create and manage Parametric Adaped Portal to create and manage Parametric studiesstudies

New workflow EditorNew workflow Editor Creation of parameterized input fileCreation of parameterized input file Manage parameter valuesManage parameter values

Workflow ManagementWorkflow Management Submit workflows by parameter rangesSubmit workflows by parameter ranges Compare outputsCompare outputs Monitor individual job statusMonitor individual job status

Pstudy ManagerPstudy Manager

VisualizationVisualization

PGRADE DemoPGRADE Demo

http://hgportal.hpcc.sztaki.hu:8080/gridsphhttp://hgportal.hpcc.sztaki.hu:8080/gridsphere/gridsphereere/gridsphere