grid computing

Nguyen Tuan AnhGrid Computing – Master Course 2008

Grid Computing

Dr. Nguyen Tuan Anh

Contacts: Email: [email protected] Tel: (84-8) 8647256, ext. 5840

Faculty of Computer Science and EngineeringHo Chi Minh city University of Technology


Course objectives

Understand “Grid” concepts

Roles of “Grid” in our modern society

Grid architectures

Understand Grid related technologies Security Web services Grid services and standards

Ways to efficiently exploit the Grid power Programming models

“Know-how” Programming the Grid


Evaluation

Self study and self research Active Efficient

Practical exercises and seminars: 50% Exercises Seminar: each group will study a Grid-related topics and present this in the

class

Final exam: 50%


References

Lecture notes http://www.cse.hcmut.edu.vn/~tuananh/courses/Grid

Books I. Foster and C. Kesselman, The Grid: Blueprint for a New Computing Infrastructure.

Morgab Kaufmann Publishers, 1999. Maozhen Li, Mark Baker, The Grid Core Technologies, Wiley, 2005. Fran Berman, Anthony J. G. Hey and Geoffrey C. Fox, Grid computing: Making the

Global Infrastructure a Reality. John Wiley & Sons Ltd, 2003.

Luis Ferrelra et al. Grid Services Programming and Application Enablement. IBM Redbooks, 2004.

Mark Endrei et al. Patterns: Service Oriented Architecture and Web Services. IBM Redbooks, 2004.

Web links: Globus project: http://www.globus.org/alliance/ Global Grid Forum: http://www.ggf.org


Grid introduction


CERN

CERN (founded 1954) = “Conseil Européen pour la

Recherche Nucléaire”

“European Organisation for Nuclear Reseach”

27 km circumference tunnel

Large

Hadron

Collider

Concorde(15 Km)

Balloon(30 Km)

CD stack with1 year LHC data!(~ 20 Km)

Mt. Blanc(4.8 Km)

50 CD-ROM

= 35 GB

6 cm


Un little bit of history …

I.Foster and C.Kesselman come from the parallel computing field Parallelism

1970-80 : The pioneers– Vectorial computers (Cray 1, Cray 2,..)

1980-90 : The exuberance Massively Parallel Computers (CrayT3D/T3E, Connection Machine,…)

1990-2000 : The profit Cluster (Beowulf, ASCII Red,..)

2000 ->? The Globalization GRID Global Computing


Overview

Grid concepts

Challenges and issues

Grid middleware

Grid applications

Grid projects

Future directions


Part I: concepts


What is “Grid”?

1001 definitions HPC researchers

Connect supercomputing sites Distributed supercomputing

Web engineers Stateful web services for accessing resources Next generation of the Internet and Web

Companies Services accessible by other users Portal to access local resources

Users Like electrical grid

…


Grid environment

11


« Grid » domains

Originally To link together supercomputing sites

Today Collaborative engineering Data exploration High-throughput computing Distributed high performance computing


A little bit of vocabulary

Usually one distinguishes three types of GRID

Knowledge GRID To share knowledge Idea: Virtual distributed laboratory

Data GRID To share the data Idea: Large scale data storage

Computing GRID To share computing power Idea: Alternative to supercomputers


The GRID today -1

Knowledge GRID Virtual distributed Institutes/Laboratories allowing remote people to collaborate as

they would be at the same location

Purpose: to share knowledge, to collaborate for researches or specific activities

Few specific tries Remote surgery ….


The GRID today - 2

Data GRID Also referenced as « peer-to-peer » technology

Versus to the « Client/Server » technology who dominated the market of distributed systems during last decades, with this approach every participant is both client and server

Nobody knows exactly where the data are The system figures to find the nearest requested data out

The requested data could be distributed on several location and same data could be located at several places (data coherence problem)

The pursuit of the data is transparent to the user Some examples:

Napster, Gnutella, pirate software, Project DATAGRID (CERN),…


The GRID today - 3

Computing GRID A supercomputer on you the table of your kitchen …

Some facts Supercomputers are very expensive and age very rapidly Except for some very specific persons, we do not need a supercomputer every day…

… but when we need one it is very frustrating not to have one ! The total time personal and other computer waste time not doing anything is huge I am (almost) permanently connected to a lot of computers. Potentially I am permanently connected to a supercomputer

First try to exploit this fact: Projet seti@home


GRID computing : different views

Virtual Supercomputing An association of several supercomputers geographically distributed (10-

1000) Each node is a parallel machine managed through a scheduling system

(batch) To provide a virtual supercomputer

Internet computing A combination of a large number of PCs

(10000 - 1000000) To exploit unused computing

Metacomputing An association of several servers providing a set of applications

Thierry Priol


“Grid” vs. cluster computing

“Partial view” vs. “Global view” of the environment

Large scale, geographically distributed

Non-structured

Heterogeneous

Dynamic, unstable

Multiple administrative domains


Supercomputer

PC

Supercomputer

My laptopMy laptop

Workstation

SMPCluster

Computing Grid: our goal

?


The dream

To make computers user friendly: To switch from a program centric view…

« Execute this program on this (these) machine(s)»

to a service centric view… « Provide me this service with this QoS level »

Today the WEB provides access to a large amount of documents and data BUT…. Access is not transparent…. URL To find the correct information is a tedious work… search engines

The GRID will change our way to use computer as many drastically

that WEB did more than 10 years ago


New characteristics

Large-scale Need for dynamic selection Partial view of the environment

Heterogeneity Hardware, OS, network, software environments (languages, libraries, tools...)

Complex unpredictable structure

Dynamic unpredictable behavior

Multiple administrative domain no centralized control


Challenges

Efficient exploitation of Grid performance New programming models Service-centric vision

Resource connectivity Middleware services Applications

Interoperability vs. performance Good interoperability -> (very) slow Fast -> not so interoperable

Grid scalability Resource management mechanisms/models

Large number of heterogeneous resources

Grid infrastructure and Grid application evaluations How to benchmark Grid systems and applications

Trust and security How to trust the environment How to protect sensitive data


Part II: Grid Middleware & components


What is Grid middleware

System software between applications and OS Provide services to applications

Discovery Execution Storage Data movements Information Service integration ... Failure detection and recovery Resource monitoring ...

Hide all complexities of the Grid environment


Grid layered architecture: principle


Grid architecture

Fabric layer Physical resources

Computers/ computing platforms Data storages Sensors Network

Layered on top of local OS

Resource & connectivity protocols layer Provide access to resources and services in the fabric layer Security management

Authentication Authorization Secure communication


Grid architecture

Collective service layers Facilitate the use of resources/services Provide meta-services

Resource discovery Resource brokers Failure detection ...

User application layers Grid applications

Parallel applications Service-oriented applications ...


Community Grid Architecture Model


Service-centric view: Grid environment

Physical Environment

Security• Authentication• Authorization• Policy implementation

Resources• Virtualization• Management• Optimization

Execution Management

• Execution Planning• Workflow• Work managers

Provisioning• Configuration• Deployment• Optimization

Data• Storage Mgmt.• Transport• Replica Mgmt.

Virtual Domains

• Service Groups• Virtual Organizations

Physical Environment

• Hardware• Network• Sensors• Equipment

I nfrastructure Profile

• Required interfaces supported by all services


Virtual organization (VO)

Grid environment can be organized as VOs VO: groups of services/resources

Similar characteristics Location/site Functions

A VO can a part of a bigger VO


Grid evolution

1996

GT1.0

Proof of concept

1999

GUSTO testbed

2002

GT3.0OGSA

2005

GT4.xWSRF

GGF

Standardization

-Basic services?-Security issues

- Grid architecture- Grid interoperability

Globus Toolkit !!!!http://www.globus.org


Globus layered architecture

Applications

Core ServicesMetacomputing

Directory Service

GRAMGlobus

Security Interface

Heartbeat Monitor

Nexus

Gloperf

Local Services

LSF

Condor MPI

NQEEasy

TCP

SolarisIrixAIX

UDP

High-level Services and Tools

POP-C++ globusrunMPI Nimrod/GProActive CC++

GlobusView Testbed Status

GASS


Globus Toolkit 4x

Sustainable changes on the services interoperability and infrastructure Open Grid Services Architecture (OGSA)

Stateful Web Services Enable the integration of user specific Grid

services Define standard interfaces

How to access Grid services Disadvantage

Slow


GT: Core service architecture


Globus Toolkit

Grid Service Specification How to write, publish and use a Grid service

GT components: GT core

Meta-services used to implement other services and service behaviors (e.g. service creation, destruction)

GT base services Use the GT core to implement Grid capacities: resource

management, information services, data transfer, etc. Other Grid services

Implemented by user to enable some enhancement capacities


GT: from other perspectives

GT containerGT security

Local-level services WS GRAM

Fork PBS LSF

GridFTP RFTP

…

VO-level services Information service…???

create

create


What GT DOES NOT address

GT focus on accessing local resources

Things still missing Coordination services

Resource/service discovery Information collection Resource connectivity Programming models/tools

Things to be improved Performance!


Part III: Grid security


Grid security: a scenario


Symmetric key


Asymmetric cryptography

UniqueMathematical relation


How to guarantee the information integrity?


Digital Signatures

A message signed by the private-key


How to identify users?

(User authentication)


Authentication models

Direct-trusted model Third-party trusted model

Sender Receiver

1: Register sender’s public key

2: store trusted public key

3: message M

4: sign(M) with sender’s private key

Key repository5: public key existed?

Sender Receiver

Trusted authority

Digital certificate


Digital certificate

Based on asymmetric cryptography

Consist of Public key User identity information (user name, organization, address, …) One or more digital signature signed by some well known Certificate

Authority (CA)


Certificate authority

Responsible for Positively identify entities requesting certificates Issuing, removing, and archiving certificates Protecting the Certificate Authority server Maintaining a namespace of unique names for certificate owners Serve signed certificates to those needing to authenticate entities Logging activity


Access the Grid


Grid authentication


Grid Delegation


Delegation


Grid authentication and delegation

Source: Ian Foster et al: A security architecture for computational Grids


Part IV: Grid applications


Computing GRID: the issue ….

Supercomputer, cluster,… How to extract the 99,999999% of the computing power of my limited powered expensive

environment

GRID environment How to extract the very power I need from the theoretically infinite powered cheap

environment

Consequence Speedup/efficiency curves are not any more relevant information..


Grid vs. Cluster computing from application view

Cluster Have applications, build a cluster for those applications High efficiency but expensive

Grid infrastructure Have existing platforms, find applications that can efficiently run on those

platforms Cheap but not well tailored to every application


Types of Grid applications

Type 1: Traditional HPC applications running within a site (VO)

Using traditional models (MPI, PVM,…) Ready-to-run, no need to modify/re-compile Role of the Grid middleware

– Resource discovery – Deploy and run the application remotely, securely on the discovered resource



Type 1: Traditional HPC applications running within a site (VO)

Using traditional models (MPI, PVM,…) Ready-to-run, no need to modify/re-compile Role of the Grid middleware

– Resource discovery – Deploy and run the application remotely, securely on the discovered

resource



Type 2: New HPC applications running across multiple sites (VOs)

Require new programming models/tools– Multiple level parallelism– Embracing parallelism

Example: bio-informatics, parameter sweeping Huge speedup can be achieved

Very few applications Role of the Grid middleware

Resource discovery Resource allocation and co-allocation Application supporting services Dynamic deployments and executions of application components


Issues

Missing high-level services QoS of resources

Heterogeneity

Code portability Binary/Byte code or source code?

Resource connectivity Firewall/NAT/ Virtual IP

Fault tolerance Resource volatility

Data protection Protect sensitive data from stealing


Part V: Grid in the world


United Kingdom

E-Science, 215M€ over 5 years e-Science will refer to the large scale science that will increasingly be carried out through distributed

global collaborations enabled by the Internet.

1 - National Network of Grid Centers

2 - Development of Generic Grid Middleware

3 - Support for e-Science Projects.

e-Science support centre

Grid Network Team

Grid Engineering Task Force

Newcastle

Edinburgh

Oxford

Glasgow

Manchester

Cardiff

Southampton

London

Belfast


Germany

D-GRID, 100M€ over 5 years (2005-2009) More than 100 academia and industry partners

Phase 1 -> 2008: Construction of a GRIDinfrastructure, development of themiddleware and a pilot to runGRID applications

Phase 2: the GRID infrastructure will beused to run e-science(enhanced science) projects

In close collaboration with IBM

Stuttgart

Leipzig

Berlin

Frankfurt

Karlsruhe Garching

Kiel

Braunschweig

Dresden

Aachen

RegensburgKaiserslau

tern Augsburg

Bielefeld

Hannover

Erlangen

Heidelberg

IlmenauWürzb

urg

Magdeburg

Marburg

Göttingen

Oldenburg

Essen

St. Augusti

n

Rostock

Global UpstreamHamb

urg

10 Gbit/s2,4 Gbit/s622 Mbit/s


Nederland

Virtual Laboratory for e-science (VL-e), 55 M€ over 5 years 21 partners in 19 institutions

The mission of the VL-E project is: To boost e-Science by the creation of an e-Science environment and doing research on methodologies.

The strategy will be: To carry out concerted research along the complete e-Science technology chain, ranging from applications to

networking, focused on new methodologies and reusable components.

The essential components of the total e-Science technology chain are: e-Science development areas, a Virtual Laboratory development area, a Large Scale Distributed computing development area, consisting of high performance networking and grid parts.


France

ACI-GRID, ~8 M€ 2001-2004 Based on « call for proposals » Use RENATER network

GRID 5000 Building a nation wide experimental platform for Grid researches (like a particle accelerator for the

computer scientists) 10/11 geographically distributed sites, every site hosts a cluster (from 256 CPUs to 1K CPUs) All sites are connected by RENATER (French Academ. Network) RENATER hosts probes to trace network condition load Design and develop a system/middleware environment for safely test and repeat experiments Only experimental platform (no production)

R eimsC aen

Rouen

Ren nes N ancy

S tras bourg

Po it ie rs

Bo rd eaux

L yon

G renob le

Tou lou se .

Mon tpellie r

Marseille

N antes

B esan çon

C lerm ont -F errand

Lim oges

P aris

D ijon

L il le

S oph ia

C om piè gne

O rlé ans

C orte

2 .4 Gbi t/s

622 Mbi t/s

155 Mbi t/s

34 M bi t/s

8 M bit/s

N R D

N IO

N O C


Europe - CERN

DATAGRID 10M€, ended beginning 2004 21 partners Feasibility project, final test bed

1000 computers, 15 Terabytes on 25 sites

Followed by…

EGEE, 4 years, 40 M€ for the first two years 70 partners in 27 countries To provide the necessary storage and computing infrastructure to LHC (and others..)


…and at HCMUT….


VN-Grid: toward a national-scale computing Grid

Main focus: infrastructure High-level services

Resource discovery and reservation Scheduling VO and policy management

OGSA and WSRF compliance Programming support

MPI POP-C++

We do not develop from scratch! Using GT for providing base services


A VN-Grid scenario

Site

Resource Discovery

ResourceAllocation/Reservation

Scheduling

Learn

in

g

Data warehouse

Monitoring


Our first prototype-to-be-built

Keep in mind the heterogeneity the dynamics “Virtual site” concept (VSite) Combine the flexibility of P2P technologies (partial view assumption) with the efficiency of

centralized management on each VSite Flexible to involve more resources Flexible security management

Multiple level authentication and authorization (VO, user,…)

Programming supports Parallel object model (POP-C++) MPI

Applications Oil exploitation (geo-physic data computation of oil fields) Supraconductor study Aviation Chip Design


VN-Grid testbed


Other projects

Factors that could lead to the success of VN-Grid SuperNode

POP-C++

EDAGrid


Supernode project

Building HPC system with programming supports

Supernode 1 (1998-2000) Supernode 2 (2004-2006)

• 4 Nodes• Configurable network topology• MPI interface

• 64 Nodes (dual Xeon)• Fast network (32 links, 1Gb/s each)• 330 GFlops effective performance• MPI interface


User Management and Billing

User

Proxy

Resource Management Job Execution

Job ManagementMonitoringScheduling

Supernode II Architechture

Remote Terminal

Request/Result

Proxy

Supernode II

Transparency!


Supernode II: tools


Object oriented application

POP-C++: Parallel Object Programming

Grid environment

ObjectObj

Object

ObjectObject

Object

• Heterogeneous• Large scale• Unstructured• Dynamic and unknown topology

• Distributed objects• Heterogeneous

execute

http://www.eif.ch/gridgroup/popc

Object

ObjectObject

Obj

Object

Object


Parallel object by extending C++

Computational environment

POP-C++ essential service abstractions

Globus Toolkit XtremWebStandalone POP-C++

GridWeb

computing

Testing Distributed

environment

Other toolkits

Other distributed

environments

POP-C++ programming

POP-C++ services for

Globus

POP-C++ services for XtremWeb

POP-C++ services for

testing

Other customizable

services

parclass Foo {

Foo() @{ od.power(100);};

async conc void MyMethod();

};

parclass Foo {

Foo() @{ od.power(100);};

async conc void MyMethod();

};

class Foo {

Foo();

void MyMethod();

};

class Foo {

Foo();

void MyMethod();

};

Foo::Foo() { ... }

void Foo::MyMethod(){...}

Foo::Foo() { ... }

void Foo::MyMethod(){...}

POP-C++C++

Shared implementation

Grid ProgrammingGrid Programming

Grid infrastructure

• Transparent, requirement driven object allocation• Various method invocation semantics

Archi

tectu

re

Archi

tectu

re


EDAGRID

Computing Grid for microchip design Synthesis and optimization

Smaller Faster More power efficient Self verification


Behavioral Specification

RTL Description

Translation

Logic Optimization

Technology Mapping

Optimized Logic Description

Unoptimized Logic Description

Microchip design processProduct Requirement

Physical Design(back-end)

Library

Constraints

GRID

MIDDLEWARE

Supercomputer, PC-Cluster

Data-storage, Sensors, Experiments

Internet, networks


Future directions

OGSA/WSRF based services

Target Grid environment Data Grid or service Grid or computing Grid?

Global services Grid topology Resource discovery Distributed information system Resource connectivity Fault tolerance

Local services Acquire resource information Security management

Grid enabled Applications Programming models Supporting tools

Nguyen Tuan AnhPage 83Grid Computing – Master Course 2008

GT: Core service architecture


Globus Toolkit

Grid Service Specification How to write, publish and use a Grid service

GT components: GT core

Meta-services used to implement other services and service behaviors (e.g. service creation, destruction)

GT base services Use the GT core to implement Grid capacities: resource management, information services, data

transfer, etc. Other Grid services

Implemented by user to enable some enhancement capacities


What GT DOES NOT address

GT focus on accessing local resources

Things still missing Coordination services

Resource/service discovery Information collection Resource connectivity Programming models/tools

Things to be improved Performance!

grid computing

Documents