Grid computing notes

By

Prof. SYED MUSTAFA, HKBK College Of Engineering

Clouds, Grids, and Clusters

Introduction:

• Grid computing is a distributed computing approach where the end userwill be ubiquitously offered any of the services of a “grid” or a network ofcomputer systems located either in a Local Area Network (LAN) or in a WideArea Network (WAN) in a spread of geographical area.

• It aims at dynamic or “runtime” selection, allocation and control ofcomputational resources such as processing power, disk storage orspecialized software systems or even data, according to the demands of theend-users.

GRID Computing – 1:

2

Clouds, Grids and Clusters- Prof. A. Syed Mustafa, HKBKCE.

UNIT – 5

Clouds, Grids, and Clusters

Text Books:

1. Anthony T. Velte, Toby J. Velte, Robert Elsenpeter: Cloud Computing, A Practical Approach, McGraw Fill, 2010.

Chapters: 1,2,3,4,5,7,8, 9,10,12,13.

2. Prabhu: Grid and Cluster Computing, PHI, 2007. Chapters:1,2,3,4,5,6,7,8,9,10,12, 13,14,15,16

Unit 1 : Text 1-Chapters-1,2

Unit 2 : Text 1-Chapters-3,4

Unit 3 : Text 1-Chapters-5,7,8

Unit 4 : Text 1-Chapters-9,10,12,13


Unit 6 : Text 2-Chapters-3,4,6



Introduction: • Grid computing also means that the end-users connect to

a grid of computing resources the way the end-users of a powersupply or water supply grid connect and draw power or water as they need,at any time and any location without any knowledge or reference to thedetails such as the exact location or nature or quantity of the resourcebeing drawn.

• IBM :▫ “A grid is a collection of distributed computing resources over a local or

wide area network, that appear to an end-user or application as one largevirtual computing system”


4


UNIT – 5

Introduction: • Vision:

Create virtual dynamic organizations through secure, coordinated resourcesharing among individuals and institutions.

• Grid Computing:An approach that spans locations, organizations, machine architectures and

software boundaries to provide unlimited power, collaboration and informationaccess to everyone connected to the grid.

The Internet is about getting computers together (connected), grid computingis about getting computers work together.

Combines the QoS of enterprise computing with ability to share theheterogeneous distributed resources – everything from applications to datastorage and servers.


5


UNIT – 5

Introduction: • Grid computing is a middleware software that manages and executes all the

activities related to:

▫ Identification of resources

▫ Allocation and deallocation of resources

▫ Consolidation of resources

• Organizations having under-utilized or over-utilized resources need a dynamically equitable distribution of resources.


6


UNIT – 5

Introduction:

The 'grid' refers to an infrastructure that enables the integrated, collaborative use of highend computers, networks, databases, and also other scientific resources includinginstruments owned and managed by various organizations.

“A computational grid is a hardware and software infrastructure that providesdependable, consistent, pervasive, and inexpensive access to high-end computationalcapabilities”


7


UNIT – 5

Introduction:

Multiple Virtual Organizations(MVO)- Collaborator agencies in a grid.

Grid is a network of computation while the Internet is network of communication.

The Grid computing Environment provides the tools and protocols for resource sharingand problem solving in dynamic, multi-institutional Virtual organizations or communitiesof user organizations.


8


UNIT – 5

Introduction:

Grid computing comprises of a combination of a decentralized architecture for resourcemanagement and a layered architecture of a specific hierarchy for the implementationof various services of the grid.

Grid Computing System can have any configuration starting with a Local AreaNetwork(LAN) or a large Wide Area Network(WAN) at the national scale or even aninternational network spanning several countries and continents.

It can span a single organization or many organizations or service providers’ space.

A grid can focus on the pooled assets of one organization or a pool f Multiple VirtualOrganizations(MVOs) all of which use common protocols so as to enable the grid to offerservices and run applications in a secure and controlled way.


9


UNIT – 5

1.1 The Data Centre:• Before the data centre concepts came, organizations maintained own servers and specialized

software.• This approach was expensive and redundant• Data Centres shared resources with organizations.• Organizations connected to a data centre may not be able to use resources from other data

centres• Concept of grid computing enables multiple data centres (same or different organizations) to be

networked and shared.• Grid is a combination of:▫ Distributed computing▫ High Performance computing▫ Disposable computing

• Grid provides a metacomputing environment, which can be a megacomputing facility for theusers.

• Grid provides computational utility to its consumers


10


UNIT – 5

1.2 Cluster Computing and Grid computing :

Clusters are aggregations of processors in parallel configurations.

• Resource allocation is performed by a centralized resource manager and scheduling system.

• A cluster comprises of multiple interconnected independent nodes that cooperatively work together as a single unified resource.

• All nodes of a cluster work cooperatively together, as a unified resource.

• Grid has resource manager for each node.

• Grid does not provide a single system view.

• Some grids are collections of clusters. Example: NSF Tera Grid


11


UNIT – 5

1.3 Metacomputing• Metacomputing is all computing and computing-oriented activity which involves

computing knowledge (science and technology) utilized for the research,development and application of different types of computing. ---- Wikipedia

• Use of powerful computing resources, transparently available to the user via anetworked environment is Metacomputing.

• Three essential steps to achieve goals of metacomputing are:

1. To integrate the large number of individual hardware and software resourcesinto a combined networked resource

2. To deploy and implement a middleware to provide a transparent view ofresources available

3. To develop and deploy optimal applications on the distributed metacomputingenvironment to take advantage of the resources.

UNIT – 5

GRID Computing – 1: Clouds, Grids and Clusters- Prof. A. Syed Mustafa,

HKBKCE.

http://en.wikipedia.org/wiki/Knowledge

http://en.wikipedia.org/wiki/Research

1.3 Metacomputing• Challenges in metacomputing –

• -Viability of the linking speeds for realistic application execution

• -ability and feasibility to execute parallely the components of an application

• Resources and originating points of data are geographically distributed – may need to processed in a distributed manner

• Metacomputing is useful when a single point usage is required for large remotely located resources.

• Metacomputing encompasses two broad categories:

• - Seamless access to high performance

• -Linking of computing resources, instruments and other resources.

UNIT – 5


HKBKCE.

1.3 Metacomputer composition

• Metacomputer is a virtual computer. It has a vrtual computing architecture.

• Its constituent components (individual servers and other computational resources) are individually not important.

• Metacomputer consists of:

1. Processors and memory

2. Network and communication software

3. Remote data access and retrieval

4. Virtual environment

UNIT – 5


HKBKCE.


1. Processors and memory

The primary resources of metacomputer is processors and the associated memory units.

A meta computer is a Single virtual view of several (large number) of processors

and their associated memory units

2. Network and communication software

Interconnected network of physically distributed processors.

Links between machines could be via modems,ISDN, Ethernet(fast/gigabit),FDDI,ATM (Asynchronous Transmission mode) or any other networkingtechnology.

High bandwidth and low latency is required to provide rapid and reliablecommunication

UNIT – 5


HKBKCE.


3. Remote data access and retrieval In a Metacomputer comprising of a large number of nodes, the data stored in the

secondary storage devices with each node is required to be accessed remotely andretrieved upon demand.

Date sizing may go upto petabytes.

Retrieval, replication and mirroring support will be required for the purposes ofrecovery and business continuity.

Ability to access remote data is a challenge in metacomputing.

Ability to manage and manipulate large quantity of remote data

4. Virtual environment

A software like an operating system, that can configure, manage and maintain meta computing environment.

UNIT – 5


HKBKCE.

1.3 Evolution of Metacomputing projects• Project FAFNER (1995) - (Factoring via Network-Enabled Recursion)

▫ Finding factors of large numbers parallely, over a large network of mathematicians who calculated the factors required for prime numbers in the context of encryption for Public Key Infrastructure(PKI).

▫ PKI is for secure communication with digital signatures.

▫ RSA( Rivest, Shamir, Adleman-Massachusetts Intstitute of Technology-MIT) Algorithm –keys were generated to factor extremely large numbers.

▫ RSA keys are 154 or 512 digit keys.

▫ Started by Bellcore Labs, Syracuse University.

▫ FAFNER was initiated to factor RSA-130 using a numerical technique called Number field Save(NFS) factorization method, using web servers for computation.

▫ To distribute the code for factorizing and related information data

UNIT – 5


HKBKCE.

1.3 Evolution of Metacomputing projects• FAFNER (1995) - (Factoring via Network-Enabled Recursion)

▫ A web interface for NFS(Number Field Sieve) was produced.

▫ Web forms with ‘get’ and ‘post’ method were used to invoke the CGI scripts from Server side written in PERL language.

▫ Single workstations with a small memory(4MB) were allowed to perform useful work using Sieve and small boundaries.

▫ A consortium of sites was deployed to run CGI scripts package locally.

▫ The monitoring was done by RSA-130 web servers hierarchically.

UNIT – 5


HKBKCE.

1.3 Evolution of Metacomputing projects

• Project I-WAY (1995) - Information Wide Area Year

▫ Experimental high-performance network, linking many servers and addressedvirtualization environments.

▫ The objective is to integrate existing high bandwidth networks with telephonesystems.

▫ The servers, datasets and software environments(for virtualization) located in 17different US locations were integrated by connecting them with 10 networks ofdifferent bandwidths and protocols using different routing and switchingtechnologies.

▫ To standardize I-WAY software interface management, key sites installed Point ofPresence(POP) computer system to serve as their respective gateways to I-WAY.

UNIT – 5


HKBKCE.



▫ These I-POP systems were UNIX workstations configured homogeneously andcontained a software environment.

▫ I- Soft which helped overcome problems and issues related to heterogeneity,scalability, security and performance.

▫ I-POP machines provided uniform authentication, resource reservation andprocess creation.

▫ I-POP also performed communication functions across the resources in I-WAY.

▫ A resource Scheduler called Computational Resource Broker(CRB) was developedor used. It consists of User-CRB, CRB-User, CRB-Local Scheduler protocols.

UNIT – 5


HKBKCE.



▫ A central scheduler maintained queues of jobs and tables indicating the state oflocal machines, allocating jobs to machines.

▫ Multiple local scheduler also operated for local scheduling.

▫ AFS file system was sued for file movement and processing functions.

▫ To support user level tools, a software called ‘Nexus’ was used to performautomating configuration mechanism.

▫ Supported Applications: Super computing, Virtual reality, multi-virtual reality, webvideo.

▫ Grid computing can be used in metacomputing mode for scientific applications.

UNIT – 5


HKBKCE.

1.4 Scientific, Business and e-Governance Grids

• Grid computing approach helped to all computing communities – businesses,scientific research and government applications.

• Scientific grids – users belong to only scientist groups

• Business grids/e-governance grids – users may belong any citizen groups usingbusiness services or government services.

• The number of users in Businesses and e-Governance are high – hence setting upsuch girds are more complex.

• Internet is the only communication network available to the citizens for accessing abusiness or an e-governance grid.

• The user interfaces, access speeds and data sizes will be large.

UNIT – 5


HKBKCE.

1.5 Web Services and Grid Computing

• Users of business and e-Governance grids will need web services over internet.

• Users of business grids will not be interested in hardware and software locations

• They are not interested in resource allocation management as well.

• Hence the need for integrating web services with grids.

• The Open Grid Services Architecture(OGSA) becomes essential to offer effective,stateful web services based on Service Oriented Architecture(SOA) on the grid.

UNIT – 5


HKBKCE.

1.6 Business computing and the Grid – a Potential Win-win situation

• Grid was initially utilized for solving very long computational problems in Scientificresearch, applications such as: weather forecasting models, molecular modelling,bioinformatics, drug design, etc.

• Computational grids have been used for several years to solve large scale problemsin basic sciences and engineering.

• By harnessing the grid approach businesses can achieve cost reduction and betterQoS.

• Grid leverages its extensive information capabilities to support the processing andstorage requirements to complete a task.

• Hence grid can provide the maximum resource utilization, providing fastest,cheapest and maximum satisfaction.

UNIT – 5


HKBKCE.


• In a Grid environment , heterogeneous computer systems located over a large areaor around the globe can be integrated and made to appear as a singlecomputational resource, to be optimally or maximally utilized by the usercommunity without any loss or wastage of time, investment or resources.

• The goal of the grid computing is to provide the users with a single view(independent of the location) and a single mechanism that can be utilized tosupport any number of computing tasks.

• The participating computer systems of a grid could be located in the same room ordistributed across the globe.

UNIT – 5


HKBKCE.


The grid computing for business is based on three factors:

1. The ability of grid to ensure more cost-effective use of a given amount of computerresources

2. A methodology to solve any difficult or large problem by using grid as a ‘largecomputer’

3. All the computing resources of a grid such as CPUs, disk storage systems andsoftware packages can be comparatively and synergistically harnessed andmanaged in collaboration towards a common business objective.

UNIT – 5


HKBKCE.

1.7 E-Governance and the Grid• In the case of e-governance, the citizens become the end user, citizen services of the

government becomes the most important and high priority application of the grid.

• citizen services are delivered as e-governance services through the web.

• The new paradigm of technology is Web services based on SOA.

• When e-governance services are delivered as Web services on Service oriented architecture, the grid has to support the Web services.

• The new grid architecture OGSA – Open grid services architecture standard offers the stateful web services for e-governance.

• Grid tools such as Globus toolkit (V4) support OGSA standard on the grid.

• It is possible to develop and offer e-governance services to citizens through web services on the grid on the OGSA standard.

UNIT – 5


HKBKCE.

Key functional requirements in Grid Computing

1. Resource Management

2. Security Management

3. Data Management

4. Services Management

UNIT – 5


HKBKCE.

Key functional requirements in Grid Computing

1. Resource ManagementThe ability to keep track, allot and remove grid resources

2. Security ManagementThe ability to ensure authenticated and authorized access to grid resources, from the users in

the external world.

3. Data ManagementThe ability of transporting , cleaning , parceling and processing the data between any two nodes in the grid, without the knowledge of the user.

4. Services ManagementThe ability of the users and applications to query and obtain response from the grid efficiently.

UNIT – 5


HKBKCE.

Six Layered Architecture of Grid ComputingLayer # Layer use

6 Application Layer Scientific/Engineering, Commercial, EGovernanceapplications

5 Middleware Layer Tools, Languages, Libraries

4 Resource Management Layer

Resource access and management, scheduling services

3 Job Management Layer

Job Scheduling, Job management, accounting

2 Security Layer Authentication and Authorization

1 Infrastructure Layer Processors, Storage, Software, Data

UNIT – 5


HKBKCE.

UNIT – 5


Six Layered Architecture of Grid Computing1. Infrastructure Layer/ Grid Fabric:

All computational resources, processors, disk storage, operating systems and software ( both Application software and System Software).

All these resources are to be monitored and dynamically allocated and deallocatedto various job requirements from the users.

This function is performed by the Middleware software(eg: Globus Toolkit).


UNIT – 5


Six Layered Architecture of Grid Computing2. Security Layer:

Performs authentication and authorization of users who intend to access the grid and its various computational resources.

3. Job Management Layer:

Part of the Core Layer of the Grid Middleware software. Designed to hide the complexities of partitioning, distributing and load balancing

of jobs. Job scheduling of various jobs.


UNIT – 5


Six Layered Architecture of Grid Computing4. Resource Management Layer:

Core Layer of Grid Management. Resource monitoring, allocation and management of the resources being drawn from

the lowest layer of the grid architecture.

5. Middleware Layer: Globus Tool Kit which comprises of tools, languages and the grid programming

Environment including Compilers and Libraries. It also includes Resource brokers- which manage the execution of the application on

distributed resources using appropriate scheduling algorithms. It uses strategies for aggregating and allocating/deallocating computational resources

in the grid. It also includes various related tools and utilities such as a grid file transfer protocol and

grid application development tools.

UNIT – 5


Six Layered Architecture of Grid Computing

6. Application Layer:

It is the layer of actual grid applications themselves: Scientific, Engineering,Commercial or EGovernance applications

Grid Architecture are now being built, based on Internet protocols and servers for thefunctionalities of communications, routing and name resolutions, etc.


UNIT – 5


Standards for Grid Computing:

Grid standards are being defined, accepted and implemented by the vendors globally.

Standards based grids are termed as third generation or 3G Grids.

The first generation(1G) grids were just local language metacomputers, with basicdistributed servers such as distributed file system, single sign-on based distributedapplications and customer communication protocols.

The 2G grids came up with an improvement with projects such as Condor and I-WAY orLegion, in all of which the underlying communication protocols software services couldbe used for developing distributed applications and services.


UNIT – 5



2G could offer very basic core/kernel and application development which requiredsignificant customization efforts.

2G Grids were not easily interoperable.

The 3G grid came with standards for interoperability.

It implements key services.


UNIT – 5



OGSA with web standards of web services such as SOAP, XML, WSDL, UDDI and thestandard of grid computing developed the project called “Globus Project ( GlobusToolkit .

Globus Project is a joint effort of Globus community of Open Source based researchersand software programmers of the whole globe, with a focus on grid research, softwaretools, test beds and applications.

OSGA become backbone of grid computing. GT4(Globus Toolkit 4) supports OGSA and WSRF.


UNIT – 5


OGSA and WSRF:

OGSA (Open Grid Services Architecture) is a distributed computing interoperabilitystandard, based on a grid service.

It leverages on web services, depending on WSDL(Web Service Description Language)interfaces( how to use the service).

The UDDI registry and WSDL document documents are used to locate the grid service. The transport protocol SOAP is used to connect data and applications for accessing a

grid service. OGSA is developed by Globus Grid Forum(CGF) with the objective of defining common,

standard and open architecture for grid based applications.


UNIT – 5


OGSA and WSRF:

OGSI (Open Grid Services Infrastructure) is the basic interoperability in terms of RPC(Remote Process Communication) for a rich set of high level services and capabilitiesthat are collectively called OGSA.

OGSA defines and describes a web services based architecture, composed of a set ofinterfaces and their behaviors, to facilitate resource sharing and accessing inheterogeneous dynamic environments.


UNIT – 5


OGSA for Resource Distribution:

OGSA is a web services based architecture, comprised of a set of interfaces and theirbehaviors, to facilitate resource sharing and accessing in heterogeneous dynamicenvironments, by relying upon the definition of a web service in WSDL, which definesmethod names, parameters and their types for grid service access.

The main theme of OGSA is, it is a service oriented grid architecture supported by gridservice, which is a special web service that provides set of well defined interfaces thatfollow specific conventions.

These interfaces of grid services all matters pertaining to service discovery, dynamicservice instance creation, life time management, notification and manageability.

All communications between the services are secure. All Services are built on Globus Toolkit. OGSA =grid structure + web services + toolkit.


UNIT – 5


OGSA for Resource Distribution:

OGSA and SOA


Grid Client

Messaging(Secure)

Grid Server(Windows Platform)

Grid Server(Linux Platform)

Grid Server(UNIX Platform)

Standard Interface, Multiple Bindings

(Java,C#)

UNIT – 5


Stateful Web Services in OGSA:

Web services Technology lacked the ‘state’ because they are stateless.

OGSA introduced ‘state’ for a web service.

OGSA defined ‘stateful’ web services using a new framework called “We ServicesResource Framework(WSRF).


UNIT – 5


WSRF ( Web Services Resource Framework) :

Web services Resource Framework (WSRF) is defined by OASIS (Organization forAdvancement of Structured Information Standards).

It specifies how we can make a Web service stateful, along with a few other usefulfeatures.

It is the result of joint effort of Web services and Grid Community.

It provides stateful Web Services that OGSA needs.

OGSA is the architecture that requires stateful Web services.

WSRF is the infrastructure on which OGSA architecture is based and built on.


UNIT – 5



Stateful Web Services architecture WSRF with OGSA


OGSA WSRF

Stateful Web services

Web Server

RequiresSpecifies

Extends

UNIT – 5



Globus Tool Kit version 4 implements both OGSA and WSRF.

Plain web services are stateless i.e past transaction details are lost.

It is necessary when web service need not to remember past transaction.

But stateful web services do remember the last transaction data.

So, have both stateless and stateful service in both hands.


UNIT – 5


WSRF ( Web Services Resource Framework) Specifications :

WSRF is a set of 5 specifications, all to be used in the management of WS-Resources.

1. WS-Resource Properties:

A resource may have 0 more resource properties.

A resource file may have 3 properties[ File name/No., Size and Description ].

It specifies clearly how the properties or attributes are defined in WSDL.

2. WS- Resource Lifetime:

Beginning and end –lifetime

It supplies information to manage lifecycle of resurces.


UNIT – 5



3. WS-Servicegroup:

It is utilized to manage a group of web services or group of WS-Resources.

Operations such as Add/remove/find a service in a group performed.

It specifies how exactly we should go about grouping the services or WS-Resources.

GT’s Features such as Index Service enable the user to group and index the servicesand also their references.

4. WS-Barefault:

It is a mechanism of enabling reporting of faults, when something goes wrongduring a web service invocation.


UNIT – 5



5. WS-Notification:

It enables a web service to be configured as a ‘notification producer’ and certainclients to be ‘notification consumers’.

It notifies all the subscribers or consumers whenever a change occurs in a WS orWS-Ref.

6. WS-Addressing:

It provides away to address web services or web resources for resource pair(WS-Resource).


UNIT – 6


Web Services and the Service Oriented Architecture (SOA) :

1. History and Background:

Before the advent of Object Oriented Programming, all programming wasmonolithic(unstructured).

Structured programming methodology came after monolithic programming.

It resulted in better modular programming methodology.

But had procedural paradigm- having procedures, subroutines and functions withglobal data maintenance- too difficult to manage.

So better and simpler abstraction –Object Oriented approach was invented.

Reusable individual modules or components can reduce the time and effort insoftware development- resulted in Component based software.


UNIT – 6




The evolution in the nature of software over the time – from Mainframe-batchprocessing was replaced with ‘online’, ‘interactive’ software.

Object- Oriented, component based software was used for online, high transactionrate, user friendly and highly available systems.

The component based, object-oriented software addressed issues of extensibility,reusability and maintainability.

Component based architecture was not adequate- so, distributed deployment,heterogeneous platform, interoperability and application integration in diverseprotocols and interfaces ware achieved.


UNIT – 6




Software Development became compartmentalized- from mainframe monolithicprogramming to client- server architecture.

Software could be broken into client side and server side software.

Client- typically a PC with GUI connected to a server at the back end with sharedresources or database tightly coupled.

From a single server LAN to Distributed systems to support WAN, to interact withmultiple WAN across the globe , internet was found.


UNIT – 6




In a large distributed system or network as Internet, the client does not know theserver and the server does not know the client.

Multi tier architecture evolved- front tier-client –user interface will talk to middletier.

Middle tier will take the request from front tier and sends it to Backend server. Thismiddle tier is known as ‘Broker’.

This 3 tier is extended as ‘n’ tier architecture over Internet.


UNIT – 6




In order to make minimal dependence between components loose coupling isneeded.

Loose coupling will help to relocate a service on to a different server withoutchanging the interface for communication with server.

Service will be discovered with its capabilities, policies, location and protocols.

Earlier to SOA and web services, technologies such as ORB(Object Resource Broker)and DCOM,COM,OLE were developed, based on CORBA(Common Object ResourceBroker) specification.


UNIT – 6




Interoperability across languages and vendors was achieved by CORBA through IIOP(Internet Inter-ORB) protocol.

IIOP was a transport protocol for distributed applications, written in either IDL orRMI.

CORBA has limitations such as parameters/return values are limited to the capabilityof representation available in the programming language.

Argument values/ return values are actual types not abstracted type.


UNIT – 6



2. Service Oriented Architecture(SOA):

SOA is based on loosely coupled ‘Service’ as an abstraction.

A Service in SOA is an exposed piece of behaviour, loosely coupled such that thecontext of service to the service is platform independent.

A Service may be dynamically discovered and called.

Dynamic discovery is possible through a directory service available at run time.

In principle, a Service an atomic operation encapsulating a particular businessprocess.


UNIT – 6




Service should be self-contained(i.e stateless).

A Service should be publishable, discoverable, communicable, self-contained, well-defined and independent of platform or other services.

Service Providers and consumers communicate via messages.

The Service contract will clearly specify the invocation procedure.

Platform independent messages are exchangeable.

The Service invocation and responses should be platform independent, i.e.interoperable across heterogeneous platforms, operated through platformindependent messages.


UNIT – 6




Web Services interact with each other automatically, without any humaninteractions, not even a programming based human intervention.

A web service(A) can interact with another web service(B) in a completelyautomated manner using the UDDI,WSDL,SOAP and XML.

Every service provider should have facility for service registration, dynamicdiscovery, encryption handling, security, interoperability independent of platformand be able to do clusters and grid computing.


UNIT – 6





Service Registry

Service Consumer

Service Provider

DiscoverRegister

Bind

Execute

UNIT – 6





UNIT – 6





UNIT – 6





UNIT – 6





UNIT – 6




Service Oriented Architecture(SOA) contains Service providers, Service consumerand Service Registry.

Initially, Service provider register themselves with Service Registry.

Then , Service consumer track and bid (engage) with Service providers.

With Heterogeneous platforms offering web services, data exchange takes placethrough XML.

Service Invocation takes place through SOAP.


• A Web service is a software system designed to support interoperable system-to-

system interaction over a network.

• “self-contained, modular business applications that have open, Internet-

oriented, standards-based interfaces” - [ UDDI consortium ]

• “A software application identified by a URI, whose interfaces and

bindings are capable of being defined, described, and discovered as

XML artifacts. A Web service supports direct interactions with other

software agents using XML-based messages exchanged via Internet-

based protocols” - [ W3C ]

Web service

64

• A Web-Service is an interface that describes a collection of operations that

are network accessible through standardized XML messaging - [ Microsoft ]

• Web services are a new breed of Web applications. They are self-contained,

self-describing, modular applications that can be published, located, and

invoked across the Web. Web services perform functions, which can be

anything from simple request to complicated business processes… Once a

Web service is deployed, other applications (and other Web services) can

discover and invoke the deployed service. [IBM]

Web service

65

• Web Services are fully XML based

• Expose interface

Clients access the service functionality through interfaces

Communication between applications as opposed to communication between users

• Self-describing modular units

• Accessible over the Web

• Interaction

XML based messages over standard Web protocols

• Registered and discoverable at the Web service registry

• Platform, Language and Protocol independence

• Compositions of Web Services are possible

66

Web Service- General characteristics

• Web Services Description Language

• Simple Object Access Protocol

• Universal Description, Discovery

and Integration

• EXtensible Markup Language

Web Service Platform

WSDL

SOAP

UDDI

XML

67

68

XML

•It is used to tag the data

•Extensible Markup Language, a specification developed by the W3C.

•It is used to create customized tags, enabling the definition, transmission,

validation and interpretation of data between applications and between

organizations.

SOAP•Communication protocol is used to transfer the data

•a lightweight xml-based messaging protocol used to encode the information in Web

service request and response messages before sending them over a network.

• SOAP messages are independent of any operating system or protocol and may be

transported using a variety of Internet protocols, including SMTP, MIME and HTTP.

http://www.webopedia.com/TERM/W/XML.html

69

WSDLWSDL is used for describing the services available

• Web Services Description Language, an XML-formatted language used to describe

a Web service's capabilities as collections of communication endpoints capable of

exchanging messages.

•WSDL is an integral part of UDDI, an XML-based worldwide business registry.

WSDL is the language that UDDI uses.

•WSDL was developed jointly by Microsoft and IBM.

UDDI•UDDI is a directory service where companies can register and search for Web

services.

•Universal Description, Discovery and Integration.

•It is a Web-based distributed directory that enables businesses to list themselves

on the Internet and discover each other

http://www.webopedia.com/TERM/W/WSDL.html

http://www.webopedia.com/TERM/U/UDDI.html

Web Service definition revisited

•A more precise definition:▫ an application component that:

Communicates via open protocols (HTTP, SMTP, etc.)

Processes XML messages framed using SOAP

Describes its messages using XML Schema

Provides an endpoint description using WSDL

Can be discovered using UDDI

Web Services Components

• XML – eXtensible Markup Language – A uniform data representation and exchange mechanism.

• SOAP – Simple Object Access Protocol – A standard way for communication.

• UDDI – Universal Description, Discovery and Integration specification –A mechanism to register and locate WS based application.

• WSDL – Web Services Description Language – A standard meta language to described the services offered.

Example – A simple Web Service

• A buyer (which might be a simple client) is ordering goods from a seller service.• The buyer finds the seller service by searching the UDDI directory. • The seller service is a Web Service whose interface is defined using Web Services

Description Language (WSDL).• The buyer is invoking the order method on the seller service using Simple Object

Access Protocol (SOAP) and the WSDL definition for the seller service.• The buyer knows what to expect in the SOAP reply message because this is defined

in the WSDL definition for the seller service.

The Web Service Model

• The Web Services architecture is based upon the interactions between three roles:▫ Service provider▫ Service registry▫ Service requestor

• The interactions involve the:▫ Publish operations▫ Find operation▫ Bind operations.

The Web Service Model (cont)

The Web Services model follows the publish, find, and bind paradigm.

1. publish 2. find

3. bind/invoke

Web Service

Registry

Web Service

ProviderWeb Service

Client

XML

• XML stands for EXtensible Markup Language. • XML is a markup language much like HTML. • XML was designed to describe data.• XML tags are not predefined. You must define your own tags.• The prefect choice for enabling cross-platform data communication in

Web Services.

XML vs HTML

An HTML example:

<html>

<body>

<h2>John Doe</h2>

<p>2 Backroads Lane<br>

New York<br>

045935435<br>

[email protected]<br>

</p>

</body>

</html>

XML vs HTML

• This will be displayed as:

• HTML specifies how the document is to be displayed, and not what information is contained in the document.

• Hard for machine to extract the embedded information. Relatively easy for human.

John Doe

2 Backroads Lane

New York

045935435

[email protected]

XML vs HTML

• Now look at the following:

• In this case:▫ The information contained is being marked, but not for

displaying. ▫ Readable by both human and machines.

<?xml version=1.0?>

<contact>

<name>John Doe</name>

<address>2 Backroads Lane</address>

<country>New York</country>

<phone>045935435</phone>

<email>[email protected]</email>

</contact>

SOAP

• SOAP originally stood for "Simple Object Access Protocol" .• Web Services expose useful functionality to Web users through a

standard Web protocol called SOAP. • Soap is an XML vocabulary standard to enable programs on separate

computers to interact across any network. SOAP is a simple markup language for describing messages between applications.

• Soap uses mainly HTTP as a transport protocol. That is, HTTP message contains a SOAP message as its payload section.

SOAP Characteristics

• SOAP has three major characteristics:

▫ Extensibility – security and WS-routing are among the extensions under development.

▫ Neutrality - SOAP can be used over any transport protocol such as HTTP, SMTP or even TCP.

▫ Independent - SOAP allows for any programming model .

SOAP Building Blocks

A SOAP message is an ordinary XML document containing the following elements: ▫ A required Envelope element that identifies the XML document as a SOAP

message.

▫ An optional Header element that contains header information.

▫ A required Body element that contains call and response information.

▫ An optional Fault element that provides information about errors that occurred while processing the message.

SOAP Request

POST /InStock HTTP/1.1

Host: www.stock.org

Content-Type: application/soap+xml; charset=utf-8 Content-Length: 150

<?xml version="1.0"?>

<soap:Envelope

xmlns:soap="http://www.w3.org/2001/12/soap-envelope"

soap:encodingStyle=http://www.w3.org/2001/12/soap-encoding”>

<soap:Body xmlns:m="http://www.stock.org/stock">

<m:GetStockPrice>

<m:StockName>IBM</m:StockName>

</m:GetStockPrice>

</soap:Body>

</soap:Envelope>

SOAP ResponseHTTP/1.1 200 OK

Content-Type: application/soap; charset=utf-8

Content-Length: 126

<?xml version="1.0"?>

<soap:Envelope xmlns:soap="http://www.w3.org/2001/12/soap-envelope"

soap:encodingStyle="http://www.w3.org/2001/12/soap-encoding">

<soap:Body xmlns:m="http://www.stock.org/stock">

<m:GetStockPriceResponse>

<m:Price>34.5</m:Price>

</m:GetStockPriceResponse>

</soap:Body>

</soap:Envelope>

SOAP Security

• SOAP uses HTTP as a transport protocol and hence can use HTTP security mainly HTTP over SSL.

• But, since SOAP can run over a number of application protocols (such as SMTP) security had to be considered.

• The WS-Security specification defines a complete encryption system.

http://msdn.microsoft.com/webservices/understanding/webservicebasics/default.aspx?pull=/library/en-us/dnglobspec/html/ws-security.asp

WSDL

• WSDL stands for Web Services Description Language.• WSDL is an XML vocabulary for describing Web services. It allows developers to

describe Web Services and their capabilities, in a standard manner. • WSDL specifies what a request message must contain and what the response

message will look like in unambiguous notation. In other words, it is a contract between the XML Web service and the client who wishes to use this service.

• In addition to describing message contents, WSDL defines where the service is available and what communications protocol is used to talk to the service.

The WSDL Document Structure

• A WSDL document is just a simple XML document.• It defines a web service using these major elements:▫ port type - The operations performed by the web service.▫ message - The messages used by the web service.▫ types - The data types used by the web service.▫ binding - The communication protocols used by the web service.

WSDL Document

<message name="GetStockPriceRequest">

<part name="stock" type="xs:string"/>

</message>

<message name="GetStockPriceResponse">

<part name="value" type="xs:string"/>

</message>

<portType name=“StocksRates">

<operation name=“GetStockPrice">

<input message=“GetStockPriceRequest"/>

<output message=“GetStockPriceResponse"/>

</operation>

</portType>

UDDI

• UDDI stands for Universal Description, Discovery and Integration.

• UDDI is a directory for storing information about web services , like yellow pages.

• UDDI is a directory of web service interfaces described by WSDL.

UNIT – 6



3. How a Web Service Works:

Initially the client interested I finding a web service in ‘Discovering Services’

( in Server 1) for a specific ‘Web Service’ (eg. Weather forecast web service).

The Discovery service in server 1 replies to the client that there is a Web Serviceon weather forecast in Server 2.

The client interacts with Server2 for Web Service for weather forecast service byasking how exactly it should interact with the service.

The service replies saying that the client should see the concerned WSDL.

The WSDL description provides the exact syntax of invocation along theparameter.


UNIT – 6





Client

Discovery Service

Web Service

Server 1

Server 2

Web Service

UNIT – 6




The client will send a SOAP message to the web server with the parameters(i.elocation).

The Web server replies in SOAP response the weather forecast for the givenlocation(parameter list by the client).

All interactions between client and server are through SOAP.


UNIT – 6



4. SOAP and WSDL:

SOAP and WSDL are the essential parts of Web services Architecture as shownbelow:

Discovery, Aggregation Processes

WSDL Description

SOAP Invocation

HTTP Transport

The transport of message is done by HTTP( Hyper Text Transfer Protocol)


UNIT – 6



4. SOAP and WSDL:

Web service invocation involves passing message between the client and theserver.

SOAP or Simple Object access Protocol specifies how the request should beformatted while being sent to the server and how the server should format itsresponse.

SOAP is the most popular choice of invocation of a web service.

HTTP is the most popular transport protocol.


UNIT – 6



5.Description:

Web services are self describable.

Once the web service is located, it can be asked to describe itself.

It describes what operations it supports and how to invoke it.

This is done by a language called ‘WSDL or Web service Descripton Language.

Web service is addressed through its URI.

Web service invocation happens automatically without human intervention.


UNIT – 6



6. Creating Web Services:

Web services are created by web service programmer using the programminglanguage of choice along with WSDL.

The existing Tools can be used to create web service and Tool will generate itselfWSDL.

SOAP code will always be generated and interpreted auto matically.

The task that a client application invokes a web service is called’ Stub’.

Based on WSDL, the tool can generate automatically the stub.

Stubs can be client side or server side stubs.


UNIT – 6




A Stub is a piece of code that generates a SOAP request from a client or interpretsa SOAP message which is a server response.

WSDL

Client Stub Server Stub

(Client side code generates (Server side code that interprets

SOAP request and also interprets SOAP request received from Clients

SOAP responses received from the and also generates SOAP responsesserver) to be sent to Clients)


UNIT – 6




Stubs are generated only once.

The stubs are reused any number of time, until replacement or modifications areintroduced to web service.

Whenever a client application requires to invoke a web service, it will call theclient side stub.

The client side Stub will generate a proper SOAP request according to the WSDL ofthe web service being called by the client side application.

This process is called ‘ marshalling or Serializing’ process.


UNIT – 6




The SOAP request is sent over a network using HTTP transport protocol.

The server receives this SOAP request from the client.

The server simply hands over the request to the stub on the server side.

The server stub will convert the SOAP request into a form which can beunderstood by the service implementation.

This step is called ‘unmarshalling or deserializing’.


UNIT – 6




After the SOAP request is deserialized, the server stub will invoke the serviceimplementation, which then executes the request it has received, generating aresult.

The result so generated by the execution of the request is handed over to theserver stub, which will now convert the result into a SOAP response.

The SOAP response is sent over the network again using HTTP transport protocol.

The Client stub receives this SOAP response and converts it into a form which canbe understood by the client application.

Finally, the application receives the result of web service invocation and uses itsresponse result.


UNIT – 6


Web Services and the Service Oriented Architecture (SOA) :7. Server Side:

Server Side Architecture


off timer, return unslept time */}

unsigned

SOAP Engine

HTTP Server

Application Server

Web Server

UNIT – 6



7. Server Side:

HTTP Server, commonly called as Web Server.

It is a piece of software which knows how to handle HTTP messages.

Eg: Apache Web server, WEA Web logic, Web sphere, Pramati, etc.

Application Server is a piece of software that hosts and provides facilities forhosting ‘applications’, which will be accessed by different clients.

The SOAP engine runs as an application inside the application server.

Eg: Jakarta Tomcat Server, Java Servlet and Java Server Pages(JSP) container that isfrequently used with Apache Axis and Globus Tool Kit.


UNIT – 6



7. Server Side:

We can operationalize web services by installing SOAP engine in applicationserver.

SOAP engine is a piece of software, which can bundle SOAP requests and alsoSOAP responses.

It is common to use a SOAP engine which can actually generate server stubs foreach individual web service.

Apache Axis is a good example of a SOAP engine.

The role of SOAP engine is limited to manipulating SOAP.


UNIT – 6



7. Server Side:

Web Sever is apiece of software that exposes a set of operations.

If the web service is implemented in Java, the web service will be a Java Class.

This web service will be requested for service from various clients through SOAPmessage, which will be analyzed by SOAP engine.


UNIT – 6

GRID Computing – 2: Web Services and the Service Oriented Architecture (SOA) :

7. Server Side:


UNIT – 6


Globus Toolkit Version 4:GT4 Architecture:


UNIT – 6


Globus Toolkit Version 4:


UNIT – 6


HKBKCE.

UNIT – 6

GRID Computing – 2: Storage Request Broker (SRB)


UNIT – 7

CLUSTER Computing – 1:

What is cluster computing?

Cluster computing is a form of computing in which a group of computersare linked together so that they can act like a single entity .

It is the technique of linking two or more computers into a network(usually through a local area network) in order to take advantage of theparallel processing power of those computers.

To meet the Grand challenge Applications (CGA), the CPU power may beincreased or alternatively, parallel computing approaches are available tobreak down the job into several components and allow multiple processorsto process them simultaneously , so that overall veryhigh performance canbe achieved.


UNIT – 7


Approach to Parallel Computing:

1. Massively parallel Processors(MPPs)

2. Symmetric Multi Processors(SMPs)

3. Cache-Coherent Non-Uniform Memory Access(CC-NUMA)

4. Distributed System Processing

5. Cluster Computing


UNIT – 7




MPP approach comprises a large parallel processing system with sharednothing architecture.

This is known as multi computer approach to parallel processing.

There may be several hundred processing elements or nodes, allconnected with each other through a high speed interconnectionnetwork/switch.

Each node will contain at least one or more processors and one mainmemory unit.


UNIT – 7




Special nodes can have additional peripherals such as disks or backupsystems connected.

Each node will have a separate copy of the operating system.

All the nodes parallelly execute the tasks assigned to them.


UNIT – 7



2. Symmetric Multi Processors(SMPs)

SMP systems can have 2 to 256/512/1028 processors.

All processors share all the global resources available, such as I/Osystems, memory and the bus.

Only a single copy of the operating system executes on all the systemsput together.


UNIT – 7



3. Cache-Coherent Non-Uniform Memory Access(CC-NUMA)

CC-NUMA is a scalable multiprocessor system architecture having acache coherent non-uniform memory access.

CC-NUMA system has a global view of all the memory.

NUMA stands for non-uniform time taken to access the nearest andfarthest part of the memory (shortest and the longest time taken toaccess the memory).


UNIT – 7



4. Distributed System Processing

Distributed Processing, with processors distributed in a network, can beunderstood as conventional networking of independent computersystems.

Each node is under its own operating system.

Each node can also be a cluster by itself.


UNIT – 7



5. Cluster Computing:

A Cluster is a collection of PCs or Workstations that are connected witheach other, using some networking technology (with an interconnectionnetwork or switch).

For high performance or parallel computing purposes a cluster will havea number of high performance PCs or workstations, connected witheach other through a high speed interconnection network.


UNIT – 7



5. Cluster Computing:

A cluster works on single, integrated collection of resources.

It can have a single system image spanning all its nodes.

In contras with a grid of distributed processing nodes, where each nodehas a separate identity, separate Operating System, separate set ofresources, which could also be shared.


UNIT – 7


How to achieve Low cost Parallel computing through clusters:

The use of a cluster of PCS and workstations to achieve highperformance computing at supercomputer scale s successful.

Clustering became practical due to the standardization of tools andutilities used by parallel applications.

Message Passing Libraries(MPL) and Data Parallel Language HPF aresome of the standards.

These standards enabled the development of tedting and debuggingapplications in a cluster( or Now-Network of Workstations/COW-Clusterof Workstations).


UNIT – 7



For testing and debugging CPU time is not required to be accounted orcharges.

For dedicated parallel platforms for regular execution, CPU time isaccounted and charged.


UNIT – 7



Components for clusters are based on the basic concepts of parallelprocessing –to use multiple processors, network interfaces, memory and harddisks with facilities as follows:

1. Multiple workstations can be connected to form a cluster to act anMPP(Massively Parallel Processor) with shared nothing architecture.

2. Memory associated with each workstations as aggregate DRAM(DynamicRAM) Cache, so as to improve virtual memory and file systemperformance.


UNIT – 7



3. RAID (Redundant Array of Inexpensive Disks)- using arrays of workstationdisks to provide cheap, highly available and scalable file storage. It usesredundantly arrays of workstations with LAN and I/O backplane. ParallelI/O is also possible with appropriate middleware.

4. Multiple Communications is possible by using many networks for paralleldata transfer between nodes.


UNIT – 7


Definition and Architecture of Cluster:

A Cluster is a specific category of parallel distributed processing systemconsisting of a collection of interconnected standalone computersworking together as a single integrated computing resource.

Each of these stand alone computers can be viewed as a node.

A node, in general, can be a single processor or multiple processorsystem, PC, workstation or SMP with Memory, I/O facilities and anOperating System.

A cluster will have 2 or more nodes together.

These nodes can be in a single cabinet or may located as stand alonenodes in a LAN.


UNIT – 7


Definition and Architecture of Cluster:

The architecture of a Cluster will have a cluster middleware whichprovides a single system view to the user.

This middleware is the essential component of the LAN or clustercabinet, which provides a single system view of the cluster.


UNIT – 7


Definition and Architecture of a Cluster:


Sequential Applications

Parallel Applications

Cluster Middleware

Interconnection network/switch

PC-1 PC-2 PC-3 PC-4 PC-n

Basic Architecture of a Cluster

UNIT – 7


Architecture of a Cluster:

In the architecture of a cluster diagram, a Cluster of PCs/Workstations isshow connected with network / switch above which a middewaresoftware operates over the cluster to provide an illusion of a singlesystem view to outside world –the users who will be runningapplications, which may be sequential and also parallel.

Parallel programming application development environment and toolssuch as parallel compilers are available.

Cluster middleware, which aims to provide Single system Image(SSI) andSystem Availability Infrastructure (SAI).


UNIT – 7



Cluster middleware, which aims to provide Single system Image(SSI) andSystem Availability Infrastructure (SAI) comprises:

1. Hardware such as memory channel.

2. Operating System kernel or gluing layer such as Solaris MC andGLUnix.

3. Application subsystem such as system management tools andelectronic forms, runtime systems such as parallel file system etc.

4. Resource management and scheduling software such as LSF or Loadsharing Facility and CODINE(Computing in Distributed NetworkEnvironment).


UNIT – 7



The network interface hardware acts as a communication processor.

This interface takes care of transmitting and receiving packets of databetween cluster nodes via a network or switch.

Fast and reliable data communication among the cluster nodes and tothe outside world is achieved by the communication software.

The nodes of the cluster work collectively as an integrated computingresource or they can operate as individual computers.


UNIT – 7


Functionality of a Cluster:

A Cluster can offer high performance, high throughput, high availability.

A cluster can be expanded, so it is scalable.

Cluster computing enables an organization to expand their processingpower using standard existing components, i.e. PCs and Workstations,which off- the-shelf commodity hardware and software available at lowcost.

The organization need not to procure proprietary hardware with highperformance/supercomputing capabilities.


UNIT – 7


Functionality of a Cluster:

The organization can conserve and preserve the existing hardware stockand assemble them into a cluster for high performance computing.

The user organization can themselves leverage on their existing hardwareto build high performance/supercomputing clusters from their ownresources – a 10 or 100 factor cost reduction in comparison with highperformance supercomputing main frame and super computers fromvendors such as Cray, IBM, SGI, NEC etc.


UNIT – 7


Categories of Cluster:

Classification of Clusters can be made on the basis of various criteria such as

1. Functionality

2. Performance

3. Availability

4. Node ownership

5. Type of node hardware

6. Node operating system

7. Node configuration

8. Level or layering of clustering


UNIT – 7



In terms of Performance, a Cluster can be classified into either HighPerformance clusters or high availability clusters.

In terms of nodes, a Cluster can be classified into dedicated clusters or non-dedicated(shared) clusters.

In case of dedicated clusters, all the nodes(PCs/workstations) are dedicatedfully for the cluster, with no independent usage.

In case of non-dedicated or shared clusters, the nodes are usedindependently by the respective end users and simultaneously CPU cyclesare stolen for cluster functionality.


UNIT – 7



Classification based on Operating System is on the basis whether the nodehas LINUX, windows, Solaris, etc.

Classification based on node configuration is based on whether the clusteris homogeneous or heterogeneous.

In a homogeneous cluster, all the nodes have similar architecture and havethe same operating system.

In a heterogeneous cluster all the nodes have different architecture and runon different operating systems.


UNIT – 7



Classification based on levels of clustering is based on the location of thenodes and their number.

In a group clusters ( number of nodes 2 to 99) nodes are connected by SAN(Storage Area Network).

Departmental clusters have nodes in 10s or 100s.

National metacomputers are WAN/ Internet based and may have severalthousand nodes(which themselves could be clusters).

International metacomputers are internet based and may have tens orhundreds of thousands or millions of nodes.


UNIT – 7



Individual clusters can be connected to form larger systems or cluster ofclusters.

Internet itself can be viewed and utilized as a cluster.


UNIT – 7


Cluster Middleware:

The essence of a cluster is to provide a Single System Image(SSI) for theentire cluster of nodes, which may be individual PCs or Workstations.

Single System Image(SSI) is the most essential aspect of a cluster ofPCs/workstations.

SSI has to be provided by a middleware, above the operating system layerand below the user interface layer.

A Cluster middleware consists of 2 sub-layers of software infrastructure:

1. SSI infrastructure 2. System Availability Infrastructure


UNIT – 7


Cluster Middleware Architecture:


SSI Middleware

User Interface

Operating System

Interconnection switch

PC-1 PC-2 PC-3 PC-4 PC-n

Architecture of a Cluster Middleware

UNIT – 7


Cluster Middleware:

SSI infrastructure attaches the operating system to all the nodes in thecluster so as to provide a single system image.

The System Availability Infrastructure aims at providing fault tolerantcomputing environment among all the nodes of a cluster by providing forfailover facilities, recovery from failures and requisite check pointing andother facilities.

Applications can be system management tools, run time systems such asparallel file systems and middleware modules such as resourcemanagement and scheduling software.


UNIT – 7


Levels and Layers of Single System Image[ Parts of SSI ]:

1. HARDWARE Layer:

SSI at hardware level allows a user to view the cluster as a shared-memory system.

A Memory Channel in a dedicated cluster environment provides virtual shared memoryamong nodes by means of intermodal space mapping.

2. OPERATING SYSTEM Kernel or ‘gluing layer’:

A Cluster operating systems will have to support the execution of both parallel applicationsand sequential applications.

The goal is to pool the resources of a cluster both parallel and sequential applications.

For this purpse ‘gang scheduling’ of parallel programs has to be done by the OS.


UNIT – 7



2. Operating System Kernel or ‘gluing layer’:

The OS has to identify the idle resources such as processors, memory and network in thecluster system and offer globalized access to all these resources.

It should achieve load balancing by supporting optimal process migration.

It should also provide fast inter process communication for the application.

OS which supports SSI are SCO UnixWare and SUN Solaris MC.

A full scale cluster wide SSI enables all physical resources and kernel resources to beaccessible from and usable by all nodes within cluster systems.

Each node should see a uniform single system image.

A full SSI at Kernel Level can be very economical, since the applications at each node willrun directly without any modifications or code rewriting.


UNIT – 7



3. Application Layer and middle/sub-system Layer:

Multiple cooperating components can be presented as a Single System Image to the useror administrator as a single application.

Single system Image can be at all levels.

A cluster administration tool offers a single point of management and control of SSIservices.

Sub-systems of cluster offer single means to create easy-to-use GUI tools.

File System SSI ensures that every node in the cluster has the same view of the data.

Global Scheduling of jobs is done by the scheduling modules in an efficient manner,offering high availability.


UNIT – 7



3. Application Layer and middle/sub-system Layer:

The benefits of SSI are well identifiable.

From any node a single image of all the resources is offered without any node for the enduser, and also the operators, to know where a particular application will run, or where aparticular resource exists.

SSI enables the system administrator to control the entire cluster as one system, with singlefamiliar interface and commands, with central system management capability.

It improves the system reliability and operational manageability.


UNIT – 7


Resource Management and Scheduling [ RMS ]:

RMS is the act of distributing the application load among the individual computer systemsin a cluster, so as to maximize the utilization throughput by maximizing the resourceutilization.

RMS is a software function and has two parts, namely resource manager and resourcescheduler.

Resource manager is responsible for all aspects of locating and allocating computationalresources, authentication and also issues relating to process creation and migration.

The resource scheduler is responsible for scheduling and queuing.

RMS is based on client-server architecture.

RMS functionality is based on Server Daemons.


UNIT – 7


Resource Management and Scheduling [ RMS ]:

Users can interact RMS through a client program or web Browser.

Applications can be online or batch mode.

The batch job is submitted after submitting its details such as location of the executablemodule and input datasets, the location where the output should be placed, system type,maximum length of the run, the nature of resources required-sequential or parallel.

RMS will execute the submitted job, based on the details submitted.


UNIT – 7

CLUSTER Computing – 1: Resource Management and Scheduling [ RMS ]:

RMS will also have to enable heterogeneous environments of PCs/workstations, and SMPand other dedicated parallel platforms to be easily and efficiently utilized with thefollowing services:

1. Load Balancing by distributing application load across all the nodes in the cluster toachieve efficient and effective use of all the resources-this may involve process migrationfrom one node to another.

2. Process migration involving termination/succession, moving and restarting processes inanother node.

3. Resource scheduling involving setting up job queues to easily manage the resources of agiven organization. Each Queue can be configured with certain attributes such as thepriority of the short jobs against the priority of the long jobs or based on specializedresources such as parallel computing platforms.


UNIT – 7

CLUSTER Computing – 1: Resource Management and Scheduling [ RMS ]:

4. Check pointing is required for managing abrupt abortions or job terminations, so thatrestart/ recovery can be easily achieved. Thus improving the system availability andreliability.

5. Better CPU utilization is achieved by RMS by engaging idle CPU cycles with jobs. In caseof idling CPUs, jobs can be assigned to them by RMS.


UNIT – 7

CLUSTER Computing – 1: Cluster Programming Environment and Tools:

Special parallel programming environments and the corresponding tools are essential forcluster computing.

They can be classifies as:

1. Thread

2. Message Passing system

3. Distributed Shared Memory (DSM) System

4. Parallel debuggers and profilers

5. Performance analysis tools

6. Cluster administration tools


UNIT – 7


1. Threads

Threads are paradigm for concurrent programming on single or multiple processorsystems.

In a conventional uniprocessor, multi processing is achieved through time sharing.

This can result in optimal utilization of resources effectively.

In multiprocessor environments, threads are primarily used to utilize all the availableprocessors.

Multithreaded applications offer quicker response to user and run faster.

Thread input creation is cheaper and easier than handling fork processes.

Threads communicate using shared variables as they are created within their parentprocess address space.


UNIT – 7


2. Message Passing System (MPS and PVM)

1. Libraries:

To write efficient parallel programs for distributed memory systems, message passinglibraries are required.

These libraries provide routines to indicate and configure messaging environment, andalso for sending and receiving packets of data.

Such system for message passing are PVM(Parallel Virtual Machine) and MPI(MessagePassing Interface).

PVM is an environment and also a library for message passing and can be used forrunning parallel applications on systems, ranging from super computers to clusters ofworkstations.


UNIT – 7



2. MPI:

It is only a standard message passing specification- it is taken by MPI forum.

MPI standard which aims at portability and efficiency, defines a message passinglibrary.

MPI and PVM libraries are available for ANSI C,C++,Fortran 77& 90,Java.


UNIT – 7



3. P-CORBA:

A Model for Parallel programming over CORBA.

It incorporates the notion of concurrency into conventional CORBA.

This model provides a way to balance the load on a CORBA based distributed system.

It provides a new idea for achieving object migration in CORBA.


UNIT – 7


3. Distributed Shared System (DSM)

It enables programming on shared variable basis and can be implemented usingsoftware or hardware solutions approach.

Software DSMs are built as separate layers on top of the communication interface.

DSM can be implemented either by only compile time methods or run time method, orcombining both.

The characteristics of hardware DSMs are: better speed than software DSMs, noburden on application and software layers.

Example of Hardware DSMs: DASH & Merdin

Example of Software DSM: Linda.


UNIT – 7


4. Parallel Debuggers

The High Performance Debugging Forum(HPDF)( under Parallel Tools ConsortiumProject) has developed HPD version specification defining functionality,semantics andsyntax for command line debugger.

A parallel debugger should be able to

manage multiple processors and threads in a process.

Display each process in a window, setup source level and machine level breakpoints.

Share break points between groups of processors.

Define and watch evolution points, display arrays and manipulate code varilables andcontents.


UNIT – 7


4. Parallel Debuggers

TotalView is a commercial debugger with GUI.

It supports parallel Programming in C, C++, Fortran 77/90, HPF, MPI & PVM.

It supports OS like Sun Solaris, UNIX and IRIX.


UNIT – 7


5. Performance Analysis Tools

It consists of

components for inserting instrumentation calls to the performance monitoring routinesinto the user’s application

a runtime performance library of monitoring routines which measure and recordvarious aspects of a program performance,

a set of tools for processing and displaying the performance data.

Commonly used tools are

Pabol- monitoring Library and analysis

AIMES- instrumentation, monitoring library analysis

MPE- Logging library and snapshot performance visualization.


UNIT – 7


6. Cluster Administration Tools

It is essential to monitor a cluster in its entirety through a GUI.

Good administrative tool means good performance of the cluster.

They are Berkely NOW(Network of Workstations),SMILE( Scalable Multi computerImplementation using Low cost Equipment).

Berkely NOW gathers and stores data in a relational database.

It uses Java applet to allow users to monitor a system from a browser.

SMILE has a tool for monitoring called K-CAP.

K-CAP uses a Java applet to connect to management node through a predefined URLaddress in the clusters.

PARMON is a comprehensive environment for monitoring large clusters.


UNIT – 7

CLUSTER Computing – 1: High Throughput Computing Cluster [ HTC ]:

HTC is an environment which provides large amounts of processing capacity to consumersover long periods of time, by fully harnessing the existing computational resourcesavailable on the cluster network.

Fro achieving high throughput computing over long periods of time, the clusteringenvironment should be robust, reliable and maintainable- surviving software andhardware failures –by appropriate failover, failback mechanisms, allowing resources tojoin and leave the cluster easily, and enabling system upgradation and reconfigurationwithout any significant down time, with very high MTBF( Mean Time Between Failure ).


UNIT – 7


HTC Cluster System-Condor:

In Condor, each user/customer is represented by a ’customer agent’ whose job is tomanage a queue of application descriptions and send the requests for resources to‘matchmaker’.

The resource ( or donor) is represented by a ‘resource agent’ whose job is to implementthe policies of resource owner i.e donor send resource offers, or donor offers resource, tothe ‘matchmaker’.

The job of the matchmaker is to find the match between resource offered by donor on theone hand and the requests for resources by customers or users on the other.

On finding a match, the matchmaker will notify both the agents that a match has beenfound.


UNIT – 7




Matchmaker

(Donor)Resource

agent

Customer agents

Request

Claiming

Protocol

Offers

Condor matchmaker

UNIT – 7



On notification, the customer agent and the resource agent perform a claiming prorocol.

Requests and offers may also contain constraints, which need to be satisfied (eg. A specificoperating system).

The matchmaker may have its own constraints.

The matchmaker may impose a specific priority policy on the customers , depending ontheir importance, urgency, nature and duration of the job, etc.

The matchmaker implement will implement a customer priority mechanism, by matchingresource requests in a specific priority order.

Customers also have a choice to break to break the allocation at any given time, dependingon their own priorities.


UNIT – 7




Layered architecture of Condor

System agent

Process Management API

Network API Workstation Statistics API

System call interface

Operating System layer

UNIT – 7



Any user or customer of Condor should be able to face and overcome several challengessuch as resources from heterogeneous environments, network protocols, remote fileaccess i.e the ability of the application to access its data files from any workstation in thecluster.

To achieve the objectives of heterogeneous resource utilization and network protocolinterface, a layered architecture is designed in Condor.

HTC system depends on the host operating systems of the nodes to provide networkcommunication, process management and workstation statistics.

As the interfaces of these functions vary from one operating system to another, the layeredarchitecture will help to overcome the difficulties involved.


UNIT – 7



HTC system management deals with APIs , which are OS dependent.

APIs for process management, network interface and workstation statistics are providedwith lower level system call interfaces to various OS environments.

The Network API provides connection-oriented, or connection-less, reliable or unreliableinterfaces.

This APIS performs all conversion between host and network integer byte ordersautomatically, checks for overflows, and also provides standard host address lookupmechanisms.

The process management API provides the ability to create, suspend, continue and kill aprocess to enable the HTC system to control the execution of a customer’s application.


UNIT – 7



When a cluster process is created, a parent process may pass to the child processes allaspects of the state such as the network connection, open files, environment variables andsecurity attributes.

The workstation API ensures that both the customer application requests and informationnecessary to implement resource owner policies are well met.

Examples of such matching will be the OS & CPU configurations, disk space requests of thecustomer to be satisfied with the owner constraints of specific timings or dates ofavailability of the resource.


UNIT – 7


How To Set Up a Simple Cluster:

A Simple Cluster can be set up by the individual user organization.

The objectives and expected performance characteristics of the cluster should be wellunderstood and defined first.

Its expected computational speed performance and latency should be decided in advance.

Once these decisions are taken, the network topology and technology suitable for meetingthe proposed cluster’s communication requirements should be identified and designed.

Technology option such as Gigabit Ethernet, optical fiber channel or ATM are chosen.

Some of the technologies compulsorily require specific topology such as switched point topoint or ring.

For Gigabit Ethernet, we can choose between direct links, hubs, switches and theirmixtures.


UNIT – 7


How To Set Up a Simple Cluster:

Hubs are to be avoided because of their latencies.

Segment switching is better than full port switching.

Direct point to point connections with crossed cabling(MDI port to MDI ports), confirming ahyper cube or a torus mesh is preferred.

Using dynamic routing protocols inside the cluster to set router automatically introducesmore traffic and congestion complexity.

Some of the operating system will provide support for bonding several physical interfacesinto a single virtual interface for higher throughput.
