karan maini - roll.no. 1

7/31/2019 Karan Maini - Roll.no. 1

1/31

PGDM (DUAL) 2004-2006

Marketing & Systems

Grid Computing

A Business Perspective

Karan MainiRoll No 1

2C

Symbiosis Institute of Management Studies


2/31

G R I D C O M P U T I N G

Acknowledgement

I take this opportunity to thank my mentor Mr. S Sasikumar for helping me research this topicand for providing a business perspective to the study.

I would also like to thank Prof. TK Ganguli

for evaluating this project.

I would like to thank Symbiosis Institute of Management Studies for providing me with aplatform to conduct this study and make it a success.

Symbiosis Institute of Management Studies 2


3/31


ABSTRACT

Getting a 54 teraflop machines computational capabilities simply by downloading a thin

client onto the machine is something that no processor can achieve solely, as of now andprobably for a long time to come. This is where technological concepts like networkcomputing, distributed computing and now grid computing have snatched the lead, simplybecause the willingness to pay, to spend more, no longer exists. As our fast-paced andmodern world swiftly develops and adapts to new and fast emerging technologies, peoplehave begun to realize that increasing processing capabilities in a single location is not asefficient as it used to be. Go ahead develop it, we need it, we are willing to pay for it this was the driving force behind chip manufacturers who tried to cram more power into asingle chip every year. That however, is history.

The realization has set in that these machines with their processing power shared are

eventually not meant for use by the common man, but instead are meant for the well beingof the common man. We see its applications in areas such as AIDS research, Genomemapping, search for other intelligent life forms, and other research related activities.

Optimum utilization of present computational power while overcoming physicalboundaries is the future. Distributed and Network computing algorithms have evolved andmoved over to give way to futuristic concepts like grid computing. Creating computingpotential equivalent to supercomputers which handle astronomical volumes of data,without the complexities of space, heat, or location, and with tremendous scope forenhancement are revolutionizing the size and complexity of problems we are and would beable to solve.

Applications of technology are incomplete till theirutility to the world at large is evident.

Grid computing is a new approach to providing a virtualised infrastructure enablingbusinesses to maximise flexibility and system response while minimising asset costs.According to research carried out by Quocirca, 52% of large European companies havelittle to no knowledge of the loading of their existing infrastructure assets, while half of theassets in those companies that do have knowledge of the loadings are heavily underutilised.

While focusing on the business impact of Grid computing, we look at what constitutes a

Grid, how Grids work and interoperate at a conceptual level, and where Grid computingoffers "quick win" gains.

Grid is a further evolution of technological changes that are already taking place - such asclustering, file sharing, Web Services, and storage and data virtualisation. Further, Gridsoffer an opportunity to minimise asset costs through greater utilisation of existing assets,and that the capability of a Grid to offer greater availability and flexibility offers distinctbusiness benefits.



4/31


Introduction

HOW DID IT START ?

Soon after the 5th generation of hardware components came about, engineers startedrealizing that despite their best efforts they could only concentrate a certain amount ofpower at a point and no more. Applying this analogy to processors they hit upon theconcept multiprocessingas the solution.

Putting more than one processor on the same motherboard not only reduced spacecomplexities but also reduced cost of integrating other components.

SMP (Symmetric Multi Processing)

A concept in which a single system with multiple processors, multiple power supplies,network interface cards and multiple storage devices, provides local fault tolerance andscalable performance.

LAN Computing

A concept where applications run in the memory space of one particular computer, whichis the server. All other terminals are dumb or rely solely on the server for their informationaccess demands. Mainframe applications are the best example of this concept.

Distributed Computing

With the advent of the client/server architecture and distributed file systems, onus ofprocessing is no longer on the central computer. Tasks can be broken down into smallerwork flows and distributed throughout the network. Results for individual work flows aresent back to the central computer and then integrated.

A task has four characteristic demands:

Networking: Delivering questions and answers

Computation: Transforming information to produce new information Database access: Access to reference information needed by the computation Database storage: Long term storage of information (needed for later access)

The ideal task is stateless (needs no database or database access), has a tiny network inputand output, and has huge computational demand. For example, a cryptographic search



5/31


problem: given the encrypted text, the clear text, and a key search range. This kind ofproblem has a few kilobytes input and output, is stateless, and can compute for days.

What is Distributed Computing?

Distributed computing is a science which solves a large problem by giving small parts

of the problem to many computers to solve and then combining the solutions for the

parts into a solution for the problem.

Distributed computing projects are designed to use the computers of hundreds of thousandsof volunteers all over the world, via the Internet, to look for extra-terrestrial radio signals,to look for prime numbers so large that they have more than ten million digits, and to findmore effective drugs to fight the AIDS virus. These projects are so large, and require somuch computing power to solve, that they would be impossible for any one computer orperson to solve in a reasonable amount of time.

Distributed computing provides an environment where one can harness idle CPU cyclesand storage space of tens, hundreds, or thousands of networked systems to work togetheron a particularly processing-intensive problem.

A number of new vendors have appeared to take advantage of the nascent market;including heavy hitters like Intel, Microsoft, Sun, and Compaq that have validated theimportance of the concept.

How It Works

In most cases today, a distributed computing architecture consists of very lightweightsoftware agents (thin clients) installed on a number of client systems, and one or morededicated distributed computing management servers. There may also be requesting clientswith software that allows them to submit jobs along with lists of their required resources.

An agent running on a processing client detects when the system is idle, notifies themanagement server that the system is available for processing, and usually requests anapplication package. The client then receives an application package from the server andruns the software when it has spare CPU cycles, and sends the results back to the server.The application may run as a screen saver, or simply in the background, without impacting

normal use of the computer. If the user of the client system needs to run his ownapplications at any time, control is immediately returned, and processing of the distributedapplication package ends. This must be essentially instantaneous, as any delay in returningcontrol will probably be unacceptable to the user.



6/31


7/31


access to which systems, and who gets priority in various situations based on rank,deadlines, and the perceived importance of each project. Obviously, robust authentication,encryption, and sandboxing are necessary to prevent unauthorized access to systems anddata within distributed systems that are meant to be inaccessible.

To take the ideal of a distributed worldwide grid to the extreme, it requires standards andprotocols for dynamic discovery and interaction of resources in diverse networkenvironments and among different distributed computing architectures. Most distributedcomputing solutions also include toolkits, libraries, and API's for porting third partyapplications to work with their platform, or for creating distributed computing applicationsfrom scratch.

What is Grid Computing ?

"A computational grid is a hardware and software infrastructure that provides

dependable, consistent, pervasive, and inexpensive access to high-end computational

capabilities".

Accessing information anytime and anywhere is the need of the hour in this new era ofmassively powerful grid-based problem solving solutions.

The explosive Grid Computing environments have now proven to be so significant thatthey are often referred to as being the world's single and most powerful computer solutions.

It has been realized that with the many benefits of Grid Computing, we have consequently

introduced both a complicated and complex global environment, which leverages amultitude of open standards and technologies in a wide variety of implementation schemes.In fact the complexity and dynamic nature of industrial problems in today's world are muchmore intensive to satisfy by the more traditional, single computational platformapproaches.

Grid Computing openly seeks and is capable of adding an infinite number of

computing devices into any grid environment, adding to the computing capability and

problem resolution tasks within the operational grid environment.

The worldwide business demand requiring intense problem-solving capabilities for

incredibly complex problems has driven in all global industry segments the need fordynamic collaboration of many ubiquitous computing resources to be able to worktogether. These difficult computational problem-solving needs have now fostered manycomplexities in virtually all computing technologies, while driving up costs and operationalaspects of the technology environments. However, this advanced computing collaborationcapability is indeed required in almost all areas of industrial and business problem solving,ranging from scientific studies to commercial solutions to academic endeavors. ForExample :



8/31


A financial organization processing wealth management applicationcollaborates with the different departments for more computational power and softwaremodeling applications. It pools a number of computing resources, which can thereby

perform faster with real-time executions of the tasks and immediate access to complexpools of data storage, all while managing complicated data transfer tasks. Thisultimately results in increased customer satisfaction with a faster turnaround time.

A group of scientists studying the atmospheric ozone layer will collecthuge amounts of experimental data, each and every day. These scientists need efficientand complex data storage capabilities across wide and geographically dispersed storagefacilities, and they need to access this data in an efficient manner based on theprocessing needs. This ultimately results in a more effective and efficient means of

performing important scientific research.

A government organization studying a natural disaster such as a chemicalspill may need to immediately collaborate with different departments in order to planfor and best manage the disaster. These organizations may need to simulate manycomputational models related to the spill in order to calculate the spread of the spill,effect of the weather on the spill, or to determine the impact on human health factors.This ultimately results in protection and safety matters being provided for public safetyissues, wildlife management and protection issues, and ecosystem protection matters:

Needles to say all of which are very key concerns.

Massive online multiplayer game scenarios for a wide community ofinternational gaming participants are occurring that require a large number of gamingcomputer servers instead of a dedicated game server. This allows international gameplayers to interact among themselves as a group in a real-time manner. This involvesthe need for on-demand allocation and provisioning of computer resources,provisioning and self-management of complex networks, and complicated data storageresources. This on-demand need is very dynamic, from moment-to-moment, and it is

always based upon the workload in the system at any given moment in time. Thisultimately results in larger gaming communities, requiring more complexinfrastructures to sustain the traffic loads, delivering more profits to the bottom lines ofgaming corporations, and higher degrees of customer satisfaction to the gamingparticipants.



9/31


Grid Computing environments must be constructed upon the

following foundations:

Coordinated resources. We should avoid building grid systems with a centralizedcontrol; instead, we must provide the necessary infrastructure for coordinationamong the resources, based on respective policies and service-level agreements.

Open standard protocols and frameworks. The use of open standards providesinteroperability and integration facilities. These standards must be applied forresource discovery, resource access, and resource coordination.

Another basic requirement of a Grid Computing system is the ability to provide the qualityof service (QoS) requirements necessary for the end-user community. These QoSvalidations must be a basic feature in any Grid system, and must be done in congruencewith the available resource matrices. These QoS features can be (for example) responsetime measures, aggregated performance, security fulfillment, resource scalability,availability, autonomic features such as event correlation and configuration management,and partial fail over mechanisms.

Data

The data aspects of any Grid Computing environment must be able to effectively manageall aspects of data, including data location, data transfer, data access, and critical aspects ofsecurity. The core functional data requirements for Grid Computing applications are:

The ability to integrate multiple distributed, heterogeneous, and independentlymanaged data sources.

The ability to provide efficient data transfer mechanisms and to provide data wherethe computation will take place for better scalability and efficiency.

The ability to provide data caching and/or replication mechanisms to minimizenetwork traffic.

The ability to provide necessary data discovery mechanisms, which allow the userto find data based on characteristics of the data.

The capability to implement data encryption and integrity checks to ensure that datais transported across the network in a secure fashion.

The ability to provide the backup/restore mechanisms and policies necessary toprevent data loss and minimize unplanned downtime across the grid.



10/31


Computation

The core functional computational requirements for grid applications are:

The ability to allow for independent management of computing resources

The ability to provide mechanisms that can intelligently and transparently selectcomputing resources capable of running a user's job

The understanding of the current and predicted loads on grid resources, resourceavailability, dynamic resource configuration, and provisioning

Failure detection and failover mechanisms Ensure appropriate security mechanisms for secure resource management, access,

and integrity

Computational and Data Grids

The quality and quantity requirements for some business-related advanced computingapplications are also becoming more and more complex. The industry is now realizing thatwe have a need, and are conducting numerous complex scientific experiments, advancedmodeling scenarios, genome matching, astronomical research, a wide variety ofsimulations, complex scientific/business modeling scenarios, and real-time personalportfolio management. These requirements can actually exceed the demands andavailability of installed computational power within an organization. Sometimes, we findthat no single organization alone satisfies some of these computational requirements.

The requirement for key data forms a core underpinning of any Grid Computing

environment. For example, in data-intensive grids, the focus is on the management of data,which is being held in a variety of data storage facilities in geographically dispersedlocations. These data sources can be databases, file systems, and storage devices. The gridsystems must also be capable of providing data virtualization services to providetransparency for data access, integration, and processing. In addition to the aboverequirements, security and privacy requirements of all respective data in a grid system isquite complex.

Service-oriented Grid architectures and building blocks

Design and develop a new service oriented grid architecture for business and industry

with special emphasis on security by the end of this decade.

Mobile grid architectures and services, merging grid technology with broadband IPnetworks, both fixed and wireless.

Design and develop Open Grid Services Architecture compliant Digital Library services.

Grid services to remotely monitor and control complex instrumentation in real-time.



11/31


12/31


Grid Benefits

Today, Grid Computing offers many solutions that already address and resolve the aboveproblems. Grid Computing solutions are constructed using a variety of technologies andopen standards. Grid Computing, in turn, provides highly scalable, highly secure, and

extremely high-performance mechanisms for discovering and negotiating access to remotecomputing resources in a seamless manner.

This makes it possible for the sharing of computing resources, on an unprecedented scale,among an infinite number of geographically distributed groups. This serves as a significanttransformation agent for individual and corporate implementations surrounding computingpractices, toward a general-purpose utility approach very similar in concept to providingelectricity or water. These electrical and water types of utilities, much like Grid Computingutilities, are available "on demand," and will always be capable of providing an always-available facility negotiated for individual or corporate utilization.

Grid Computing systems are being applied in several important scientific research andcollaboration projects; however, this does not preclude the importance of Grid Computingin business-, academic-, and industry-related fields. The commercialization of GridComputing invites and addresses a key architectural alignment with several existingcommercial frameworks for improved interoperability and integration.

Increasing pressure on development and research costs, faster time-to-market, greaterthroughput, and improved quality and innovation are always foremost in the minds ofadministrators - while computational needs are outpacing the ability of organizations todeploy sufficient resources to meet growing workload demands.

Grid Computing delivers on the potential in the growth and abundance of networkconnected systems and bandwidth: computation, collaboration and communication over theAdvanced Web. At the heart of Grid Computing is a computing infrastructure that providesdependable, consistent, pervasive and inexpensive access to computational capabilities. Bypooling federated assets into a virtual system, a grid provides a single point of access topowerful distributed resources.

Researchers working to solve many of the most difficult scientific problems have longunderstood the potential of such shared distributed computing systems. Development teamsfocused on technical products, like semiconductors, are using Grid Computing to achievehigher throughput. Likewise, the business community is beginning to recognize the

importance of distributed systems in applications such as data mining and economicmodeling.

With a grid, networked resources -- desktops, servers, storage, databases, and evenscientific instruments -- can be combined to deploy massive computing power whereverand whenever it is needed most. Users can find resources quickly, use them efficiently, andscale them seamlessly.



13/31


Using Grids as a Tool for Resource Sharing

Grids can be categorized by the type of resource that is shared, and grid projects, systems,

and applications can fall into one or more categories.

Compute Grids The original concept for grid technology arose from the need to providehigh-performance computing on demand to the academic and research communities. Gridinfrastructures were developed to deliver supercomputing power to large scientific projectswith complex number-crunching applications that required a vast number of calculations.Compute grids provide high- throughput computing through the coordinated use of manycomputers that are usually geographically distributed.

In a compute grid, a single application is split into smaller pieces to run on many differentcomputers simultaneously, producing supercomputer speed from off-the-shelf hardware.

Compute grids can significantly improve the speed and efficiency of executing applicationsthat involve complex and compute- intensive modeling, simulations, and animations.

Compute grids enable an organization to harvest spare cycles on servers, workstations, ordesktops, or on some combination of these. In many enterprise grid projects that areinitiated by the IT department, the interest is primarily in making optimal use of servercycle rather than PC cycles, since the server environment is generally more controlled andthe IT department perceives less business risk. To date, the vast majority of grid computingresearch and commercial product development has focused on compute grids.

Data Grids Data grids involve accessing geographically distributed data. The term

"information grid" is sometimes used synonymously with data grid. Data grids take gridcomputing to the next level, beyond a means of increasing computing power to a means ofcollaborating and sharing data and information resources. As a result, data grids canfacilitate collaboration while protecting valuable intellectual property.

The need to share massive amounts of data as well as computing resources across manylocations is typical of many academic and research efforts. Enterprises increasingly need toshare large files and data sets across multiple locations as well. For example, a typicalpharmaceutical company has research teams around the world that must share data. Mostenterprises solve the problem of sharing data access across a wide area through the use ofmanual processes and tools such as file transfer protocol (FTP). Data grids offer the

promise of providing simplified data access across multiple locations and systems, byproviding distributed management of large quantities of data.

Instrumentation Grids Instrumentation grids provide shared access to expensive and/orunique scientific instruments such as radio telescopes and electron microscopes for nearreal-time data processing and analysis. To date, instrumentation grids are of peculiarinterest to the research and academic communities. As life sciences and manufacturing



14/31


processes of all kinds become more highly automated, however, there should be increasedinterest within IT departments and enterprises.

Application Grids Application grids are the least well-defined and developed of the fourresource classes of grids. In concept, they ensure secure, wide- area application access and

utilization. In today's IT world, application grids hold significant promise of decreasing thecomplexity of developing and implementing cross-enterprise and multi-enterpriseapplications.

It is likely that once Web services become more wide-spread as a method for applicationdevelopment and interoperability, the resulting disaggregation of application componentswill set the stage for the emergence of application grids.

Enabling New Business Models

A number of technology transitions are taking place or will take place within the next fiveyears that will lower the barriers that exist today to deploy, maintain, and run applicationson computer grids.

The grid is not only of interest to scientists and engineers running applicationsthat is thetraditional user community for grids. Grid deployments will encompass a broad swath ofindustry verticals that will take the grid well beyond its High Performance Computing(HPC) roots. Beyond capabilities delivered to end users, every participant in the ecosystemhas a vested interest in the acceleration in grid uptake: users enjoying new and powerfulcapabilities, vendors seeking new channels and additional revenues, and organizations

discovering that grid deployment can bring associated cost reductions and a welcomecompetitive edge.

While attempts at predicting discontinuous events are not usually very accurate atdetermining actual outcome, the authors believe that the process of building a thoughtexperiment is intrinsically useful. Moreover, the readers, far from being mere witnesses,will find that these ideas will bring other powerful ideas by association that will lead to apositive influence when it comes to grid evolution.

Hardware Configurations: Nodes, Clusters, and Grids

A simple three-level abstraction to describe the following grid hardware:

NodesA computer in the traditional sense: a desktop or laptop personal computer (PC),or a server in any incarnation, including a self-standing pedestal, a rack module, or ablade, containing one or more central processing units (CPUs) in a SymmetricMultiprocessor (SMP), NUMA, or Cache Coherent Non-Uniform Memory Access(ccNUMA) configuration.

http://www.intel.com/business/bss/swapps/hpc/http://www.intel.com/business/bss/swapps/hpc/


15/31


ClusterA collection of related nodes.

GridA collection of clusters.

The nodes in a cluster are connected via some fast interconnect technology. Before theintroduction of InfiniBand and PCI Express technologies, there was a tradeoff between arelatively high-performance, single-sourced, expensive technology and an economical,standards-based, but lower performance technology. Ethernet, a technology designed fornetworking, is commonly used in cost-constrained clusters. This setup introducesbottlenecks in parallel applications that require tight node-to-node coordination. Theadoption of InfiniBand-based interconnects promises to remove this tradeoff.

The clusters in a grid can be connected via local area network (LAN) technology,constituting an intra-gridthat is a grid deployed within departmental boundariesorconnected by wide area network (WAN) technology, constituting an inter-grid that can

span the whole globe.

This model includes boundary cases as particular instances: a grid consisting of exactly onecluster is exemplified by a cluster accessible to a large community, front ended with gridmiddleware. Through Web services technology, users in a HPC shop can submit jobs forexecution through a single, local interface, not even realizing that the job may end up beingexecuted thousands of miles away. In this way, it is possible for the supporting informationtechnology (IT) department to optimize costs across a number of facilities around theworld, including outsourced service providers.

Conversely, a large clustereven one that contains thousands of nodesmay not be a grid

if it does not have the infrastructure and processes that characterize a grid. Remote accessmay need to be accomplished through relatively limited operating system (OS) utilitiessuch as rlogin or telnet or through customized Web interfaces.

A grid made up of single nodes defaults to the setup used in cycle scavenging, which isdiscussed in the full white paper[PDF 583KB] this article is derived from.

This three-tier node-cluster-grid model encompasses grids of greater complexity throughrecursion: grids of grids are possible, including grids with functional specialization. Thisfunctional specialization can happen at the lower levels for technical reasons (for example,a grid might consist of nodes of a certain memory size) or for economic reasons (for

example, a grid might be deployed at a certain geographical location because of costconsiderations).

Business Advantages that Drive Grid Adoption

ftp://download.intel.com/business/bss/technologies/grid/grid_computing.pdfftp://download.intel.com/business/bss/technologies/grid/grid_computing.pdf


16/31


A grid is essentially a set of computing resources shared over a network. Grids differ frommore traditional distributed systems, such as the classic n-tier systems, in the way itsresources are utilized. In a conventional environment, resources are dedicated: a PC orlaptop has an owner, and a server supports a specific application. A grid becomes usefuland meaningful when it both encompasses a large set of resources and serves a sizable

community.

The large set of resources associated with a grid makes it attractive to users in spite of theoverhead (and the complexity) of sharing the resource, and the grid infrastructure allowsthe investment to be shared over a large community. If the grid were an exclusive resource,it would have to be a lot smaller for the same level of investment.

In a grid environment, the binding between an application and the host on which it runsbegins to blur: the execution of a long-running program can be allocated to multiplemachines to reduce the time (also known as wall clock time or actual time) that it takes torun the application. Generally, a program designed to run in parallel will take less time to

run as more nodes are added, until algorithmic or physical bottlenecks develop or until theaccount limits are reached.

Two assumptions must hold for an application to take advantage of a grid:

Applications need to be re-engineered to scale up and down in this environment.

The system must support the dynamic resource allocation as called by applications.

As technology advances, it will become easier to attain both of these conditions, althoughmost commercial applications today cannot satisfy either of them without extensiveretrofitting.

Shared HeterogeneityA Transportation Analogy

Transportation systems follow a similar philosophy as grids, in terms of making large-scaleresources available to users on a shared basis. Jet aircraft may cost anywhere between $50and $200 million. A private aircraft might provide excellent service to its owner on a coast-to-coast flight. The obvious shortcoming of this solution, however, is that the cost of the

plane and the fuel it takes to fly it across the continent are out of reach for most people, andin any case, it probably does not represent the best use of capital for general-purposetransportation. The reason why millions of passengers can travel like this every year isbecause aircraft resources are sharedand any single user pays only for the seats usednot for a complete jet and the infrastructure behind it.

Shared-resource models come with overheads: users need to make reservations andmanage their time to predetermined schedules, and they must wait in line to get a seat. The



17/31


actual route may not be optimized for point-to-point performance: passengers may have totransfer through a hub, and the departure and destination airports may not be convenientrelative to the passenger's travel plans, requiring additional hops or some travel in a car.

Aircraft used for shared transportation are architected for this purpose. Aircraft designed

for personal transportation are significantly smaller, and would not be very efficient as ashared resource.

Transportation systems are also heterogeneous, where sharing exists on a continuum. In anair-transportation system, users choose among a variety of dedicated resources, includinggeneral aviation, executive aircraft, time-shared aircraft, commuter aircraft, and the verylarge aircraft used in long-haul flights. Likewise, grids tend to gravitate towardheterogeneity in equipment availability during their lifetime, with nodes going throughincremental upgrades. Grids tend to be deployed under diverse business models.

While the air-transportation system is an instructive instantiation of a grid, it is so

embedded in the fabric of society that we scarcely consider it as such. Computer systemswill likely evolve in a similar way as aviation did 60 years agogradually gravitatingtoward an environment of networked, shared resources as technology and processesimprove.

Fungibility and Virtualization in Grids

Ideally, the resources in a computing grid should be fungible and virtualized. Two

resources in a system are fungible if one can be used instead of the other with no loss offunctionality. Two single dollar bills are fungible, in the sense that they will each purchasethe same amount of goods, even if one is destroyed. In contrast, in most computer systemstoday, if one of two physically identical servers breaks, the second is not likely to be ableto take over smoothly. The second server may not be in the right place, or the brokenserver may contain critical data on one of its hard drives, without which the computationcannot continue.

A system can be architected to attain fungibility, for instance, by keeping data separatefrom the servers that process it. A long-running computation can checkpoint its data everyso often, so that if a host breaks, the new host can, when it comes online, pick up the

computation at the last checkpoint when it comes online. If the server was running anenterprise application, it could unwind any uncommitted transactions and proceed fromthere. An online user may notice a hiccup, but the computations are correct.

A virtualized resource has been abstracted out of certain physical limitations. For instance,any 32-bit program can access a 4 GB memory virtual space, even if the amount of actualphysical memory is substantially less. Virtualization can also apply to whole machines:multiple logical servers can be created out of a single physical server. These logical servers



18/31


run their own copies of the operating system and applications. This setup makes sense in aconsolidation setting, where the cost of maintaining the consolidated server is less than itwould cost if the machines were hosted in separate, smaller machines. A hosting serviceprovider can provide a client with what looks like an isolated machine but which is actuallya virtualized portion of a larger machine.

The nodes in a cluster may be "heavy" in the sense of being built as two, four, or moreCPUs sharing memory in an SMP configuration. Programs that take more than one node torun can operate in a hybrid Message Passing Interface (MPI)/OpenMP configuration.These programs expose large-grain parallelism, with major portions running in differentnodes using the MPI message-passing library. Within one node, each portion is split into anumber of threads that are allocated to the CPUs within a node. Building software to ahybrid configuration can increase development costs enormously.

Fungibility helps improve operational behaviors. A node operating in a fungible fashioncan be taken out of operation and replaced by another one on the fly. In a lights-out

environment, malfunctioning nodes can be left in the rack until the next scheduledmaintenance.

In a highly virtualized, fungible, and modularized environment, deploying computingresources in small increments to respond to correspondingly small variations in demand ispossible. Contrast this to the mainframe environment two decades ago: because of theexpense involved, a shop would wait until the resources of an existing mainframe weremaxed out before purchasing and bringing in a new one in what was literally a forkliftupgrade.

The main innovation brought up by IBM's System/360 was the ability to run the same

software base over a range of machine sizes. An organization could purchase a biggermachine as business grew. This change was expected to happen over months or years. Thiscapability represented enormous progress over having to reimplement the application basefor every new model, as the case was before.

The bar for business agility today is much higher. The expectation for the grid is thatresources dedicated to applications can be scaled up and down almost in real-time.Outsourcing to service providers represents an alternative over long procurement cycles.Because commodity servers are less expensive than mainframes, the budgetary impact ofadding a new server is much smaller than adding or upgrading a mainframe. Despite thisaffordability, however, not all applications can take advantage of extra servers smoothly.

The capability for incremental deployment simplifies business processes and reduces thecost of doing business. It enables new business models, such as utility computing, whereservice provisioning is metered to match demand.

A pure utility model is not yet practical today, because the concept can be taken only sofar. Even traditional utilities have different granularities and costs. Consider, for example, atraditional electric utility company, where electrons have different costs depending on the



19/31


time of day and the energy source with which they were generated. Most utilities hide thisfact, presenting most residential customers with a single, integrated bill. On-demandcomputing is a more attainable degree of utility computing, where relatively nonfungibleresources are allocated dynamically, within certain restrictions. One example is capacity-on-demand, where a large server is sold with extra CPUs that are turned on at customer

request. A restriction is that the new CPUs cannot be turned off, and hence the rates cannotbe rolled back.

Implementation

Practical grids are generally described in terms of layers. The lowest layers (theplatform) comprise the hardware resources, including computers, networks, databases,instruments, and interface devices. These devices, which will be geographically distributed,may present their data in very different formats, are likely to have different qualities ofservice (e.g. communication speeds, bandwidth) and are likely to utilize different operatingsystems and processor architectures. A key concept is that the hardware resources canchange over time - some may be withdrawn, upgraded or replaced by newer models, othersmay change their performance to adapt to local conditions - for example restrictions in theavailable communications bandwidth.

The middle layers (referred to as middleware) provide a set of software functions that

buffer the user from administrative tasks associated with access to the disparateresources. These functions are made available as services and some provide a jacketaround the hardware interfaces, such that the different hardware platforms present a unifiedinterface to different applications. Other functions manage the underlying fabric, such asidentification and scheduling of resources in a secure and auditable way. The middle layeralso provides the ability to make frequently used patterns of functions available as acomposed higher-level service using workflow techniques.



20/31


The highest layers contain the userapplication services. Pilot projects have already beencarried out in user application areas, such as life sciences (e.g. computational biology,genomics), engineering (e.g. simulation and modeling, just in time maintenance) andhealthcare (e.g. diagnosis, telematics).

These services could include horizontal functions such as workflow (the linkage ofmultiple services into a single service), web portals, data visualization and thelanguage/semantic concepts appropriate to different application sectors.

Grid Developments and Deployment

There is emerging evidence that the technology can achieve significant operational benefits(e.g. in telemedicine), improvements in performance (e.g. in climate modeling andgenomics) and a significant reduction in costs. Nevertheless, current grid technologies arenot yet viewed as sufficiently mature for industry scale use, and remain largely unproven interms of security, reliability, scalability, and performance.

Short term

For the short term, Gridis most likely to be introduced into large organizations as internalEnterprise grids, i.e. built behind firewalls and used within a limited trust domain,perhaps with controlled links to external grids. A good analogy would be the adoption intobusiness of the Internet, where the first step was often the roll out of a secure internalcompany Intranet, with a gradual extension of capabilities (and hence opportunity formisuse) towards fully ubiquitous Internet access.

Centralized management is expected to be the only way to guarantee qualities of service.

Typically users of this early technology will be expecting to achieve IT cost reduction,increased efficiency, some innovation and flexibility in business processes. At the sametime the distinction between web services and grid services is expected to disappear, withthe capabilities of one merging into the other and the interoperability between the twostandards being taken for granted.

Medium Term

In the mid term, we can expect see wider adoption - largely for resource virtualization andmass access. The technology will be particularly appropriate for applications that utilizebroadband and mobile/air interfaces, such as on-line gaming, visualization-on-demand

and applied industrial research. The emphasis will move from use within a singleorganization to use across organizational domains and within Virtual Organizations,requiring issues such as ownership, management and accounting to be handled withintrusted partnerships. There will be a shift in value from provision of compute power toprovision of information and knowledge.

At the same time open standards based tooling for building service oriented applicationsare likely to emerge and Grid technology will start to be incorporated into off-the-shelf



21/31


products. This will lead to standard consumer access to virtualized compute and dataresources, enabling a whole new range of consumer services to be delivered.

Long term

In the longer term, Gridis likely to become a prerequisite for business success - central tobusiness processes, new types of service, and a central component of product developmentand customer solutions. A key business change will be the establishment of trusted serviceproviders, probably acting on a global scale and disrupting the current supply chains andregulatory environments.

Therefore :

The Grid -- the IT infrastructure of the future -- promises to transform computation,communication, and collaboration. Over time, these will be seen in the context of

grids -- academic grids, enterprise grids, research grids, entertainment grids,community grids, and so on. Grids will become service-driven with lightweightclients accessing computing resources over the Internet. Datacenters will be safe,reliable, and available from anywhere in the world. Applications will be part of awide spectrum of network-delivered services that include compute cycles, dataprocessing tools, accounting and monitoring, and more.

Grid computing and related technologies will only be adopted by commercial usersif they are confident that their data and privacy can be adequately protected and thatthe Grid will be at least as scaleable, robust and reliable as their own in-house ITsystems. Thus, new Internet technologies and standards such as IPv6 take on evengreater importance. Needless to say, users of the Grid want easy, affordable,ubiquitous, broadband access to the Internet.

Similar to the public policy issues raise by the development of electronic commerceand electronic government, Grids raise a number of public policy issues: dataprivacy, information and cyber security, liability, antitrust, intellectual property,access, taxes, tariffs, as well as usage for education, government, and regionaldevelopment.

Distributed vs Grid Computing

Distributed computing is a subset of grid computing.



22/31


Grid Computing got its name because it strives for an ideal scenario in which the CPUcycles and storage of millions of systems across a worldwide network function as aflexible, readily accessible pool that could be harnessed by anyone who needs it, similar tothe way power companies and their users share the electrical grid.

Sun defines a computational grid as "a hardware and software infrastructure that providesdependable, consistent, pervasive, and inexpensive access to computational capabilities."Grid computing can encompass desktop PCs, but more often than not its focus is on morepowerful workstations, servers, and even mainframes and supercomputers working onproblems involving huge datasets that can run for days. And grid computing leans more todedicated systems, than systems primarily used for other tasks.

Large-scale distributed computing usually refers to a similar concept, but is more gearedto pooling the resources of hundreds or thousands of networked end-user PCs, whichindividually are more limited in their memory and processing power, and whose primarypurpose is not distributed computing, but rather serving their user. There are various levels

and types of distributed computing architectures, and both Grid and distributed computingdon't have to be implemented on a massive scale. They can be limited to CPUs among agroup

SOME APPLICATIONS OF DISTRIBUTED/GRID

COMPUTING

Render-farms for making animated movies

Rendering a frame can take many CPU hours, so a Grid-scale render farm begins tomake sense. For example, Pixar's Toy Story 2 images are very CPU intensive - a 200MB image can take several CPU hours to render. The instruction density was 200k to600k instructions per byte. This could be structured as a grid computation - sending a50MB task to a server that computes for ten hours and returns a 200MB image.

Science

Search for extra-terrestrial radio signals at SETI@home

SETI@Home is the largest public distributed computing project in terms of computingpower.



23/31


On September 26, 2001 it reached the ZettaFLOP (1021 floating point operations)mark--a new world record--performing calculations at an average of 71TeraFLOPs/second.

For comparison, the fastest individual computer in the world is IBM's ASCI White,

which runs at 12.3 TeraFLOPs/second.

On June 1, 2002, the project completed over 1 million CPU years of computation. OnAugust 19, 2003, the project processed its 1 billionth work unit.

As of June 14, 2002, the project has found 3.2 billion spikes and 266 million Gaussians.

Evolution@home

A grand-challenge computation research program to study evolution. The first simulator forthe project "helps uncover potential genetic causes of extinction for endangered andnot-yet-endangered species by investigating Mullers Ratchet. Improving andUnderstanding of such genomic decay might one day be used to fight it."

As of October 24, 2002, more than 16.3 years of CPU time have been contributed to theproject.

Climateprediction.net

Predicts Earth's climate 50 years from now. The project uses a large-scale Monte Carlosimulation to predict what the climate will do in the future.



24/31


On June 22, 2004, the project began a new phase of its experiment, to study "THCslowdown," or how climate might change as CO2 changes in the event of a decrease in thestrength of the thermohaline circulation.The project client has some large requirements. In particular, one work unit takes up to 6weeks to complete on a 1.4 GHz CPU.

Distributed Particle Accelerator Design

Helps design a more efficient particle accelerator Distributed Particle Accelerator Designproject. The project "simulates the pion-to-muon decay channel (grey cylinderssurrounding a straight blue path) of the RAL Neutrino Factory front end design. This isdifferent to the previous versions of the solenoid-channel-only optimisation because it

varies all parameters of the solenoids independently of one another.

Here the client does not need to contact a project server to get work. It submits results viaftp whenever it accumulates more than 100 Kbytes of results.

Helps "assemble a powerful, predictive electronic atlas of Earth's biologicaldiversity".

Participants "compute, map and provide knowledge of" where Earth's species of plants andanimals live currently, where they could potentially live, and where and how they couldspread across different regions of the world.

Results of the project are used "for biodiversity research, education and conservationworldwide, especially to forecast environmental events and inform public policy withleading-edge science."



25/31


Helps design the next generation ofself-diagnosing computer circuits in the DistributedHardware Evolution Project. The project client evolves populations of individualcomputer circuits with Built-In Self-Test (BIST, a way for a circuit to detect whether it isproducing results correctly) and then migrates the circuits to other project clients tocompete with their circuit populations. Self-diagnosing circuits are important to mission-critical systems exposed to radiation.

"As an increasing number of mission critical tasks are automated, self-checking circuits areof paramount importance. For example in medical applications (heart monitors,pacemakers), transport (aeroplane hardware, traffic lights, car ABS braking), space(satellites, probes) and industrial facilites (nuclear power plants) and more to come in the

future as cars start driving themselves, surgical operations are performed remotely, etc.. Inall these areas human lives or great economic loss are at risk.

The project uses Genetic Algorithms and Evolutionary Strategies to design improvedcircuits.

Company Specific Initiatives

ORACLE

It promises not only to change the way you run your data center, but to change the way you think aboutthe data center itself. It adapts to your changing business needs so that you can spend more time thinkingabout how to run your business, knowing that your infrastructure will respond with the reliable, secureperformance your applications need. It represents a significant rethinking of the traditional role of softwareinfrastructure in areas such as system performance, clustering and storage. The software is the firstinfrastructure designed for Grid computing

] Larry Ellison

Oracle Database 10g, the first relational database designed for Grid Computing,information is securely consolidated and always available.

Oracle Database 10ghas the lowest total cost of ownership by making the most efficientuse of hardware and IT resources.



26/31


EA GAMES

Electronic Arts proves a Grid Built with Oracle9 i Technology on Commodity Clusters Providesthe Most Cost-effective Database Solution

"Oracle allowed us to handle a much higher level of complexity at a much higher

performance threshold at a lower price point than previous technology."

-- Marc West, Senior Vice President and Worldwide

Chief Information Officer, Electronic Arts Inc.

University of Tennessee

The University of Tennessee is deeply involved in grid computing research, and to this end HPdeveloped the Scalable Intracampus Research Grid (SInRG) project.

SInRG's vision is one in which a massive pool of distributed computing resources becomes aroutine and seamlessly integrated part of the normal research-computing environment for verylarge communities of users.

Westgrid

WestGrid is a $44 million capital project, supported by another $4 million in operating costs, topurchase and install an innovative computing infrastructure across BC and Alberta over the nexttwo years. It is designed to make powerful computing facilities for both computation and visuallyrich collaboration available to researchers.

Sun Microsystems

Sun is changing the very nature of utility computing with the new Sun Grid utility offerings,enabling you to purchase computing power as you need it, without the long-term lifecycle costsrelated to capital, management, depreciation, and floor space. Sun Grid radically simplifies theway you select, acquire, and use next generation IT infrastructure.

Sun Grid makes complex technology simple to use via a single point of contact be it a desktop,a call center, or an enterprise. Sun Grid allows you to derive immediate productivity andeconomic benefits from our open, grid-based computing infrastructure. This utility model givesyou more choice and control on how you purchase and leverage IT power for competitiveadvantage.



27/31


Commercial Aspects

Grid computing provides consistent, inexpensive access to computational resources(supercomputers, storage systems, data sources, instruments, and people) regardless of their

physical location or access point. As such, The Grid provides a single, unified resource forsolving large-scale compute and data intensive computing applications.

Grid enables the selection, aggregation, and sharing of information resources resident in multipleadministrative domains and across geographic areas. These information resources are sharedbased upon their availability, capability, and cost, as well as the users quality of service (QoS)requirements.

Grid computing is meant to:

Reduce total cost of ownership (TCO) Aggregate and improve efficiency of computing, data, and storage resources Enable the creation of virtual organizations for applications anddata sharing No long-term contracts

Simple/standard Price transparency

High performance Secure

Eco-friendly



28/31


This new paradigm will enable heterogeneous computing resources of all kinds to be shared overnetworks and reallocated dynamically across applications, processes, and users to a greater degreethan ever before possible. It will give office and even home machines the ability to reach intocyberspace, find resources wherever they may be, and assemble them on the fly into whatever

applications are needed. In this respect, grid computing is a key foundational technology of thisnew paradigm.

There are many industries where grid computing has applicability. In the financial servicesindustry, there is no end to the consumption of CPUs. These companies need to have their riskmanagement applications executed as quickly as possible in order to be competitive. In light ofthe energy crisis in the oil and gas industry, there are many energy companies using gridcomputing for seismic processing and reservoir simulations. Companies can use additional CPUcapacity to expedite testing and get their products to market ahead of the competition. Automanufacturers can get better safety ratings with more comprehensive crash testing. On a daily

basis, banks can be more risk-averse in their trades by drawing on more extensive data. Otherindustries where grid computing is applicable include media/entertainment, manufacturing,government/education, health sciences, and information sciences.

Billing

To get to a true utility computing model, one will have to employ a billing model where one cansomehow work out the per unit of application, memory, network bandwidth, CPU, and databasecost and so on. And then present it in some kind of aggregated unit cost. Most organisations thathave a grid implementation use a very basic billing model. Even in outsourcing contracts thathave been around for many years, most of them are billing based on per user or per CPU if its

purely compute power.

Grid commercial exploitation

Grid community has been traditionally too academic with lack of focus on commercialization

Projects did not start from business models but from technologies.

10-15 years of history, scientific up to few years ago

Academia has the knowledge that have to be transferred to industry for exploitation

GRID more promising now:

Grasp the market opportunity Pervasive networks-pervasive grid, convergence with web services standards

Grid is no longer limited to High Performance Computing: we are facing soon the NextGeneration Grid (NGG)

It is necessary a radical advance in pervasiveness of grid computing. It is necessary a leveragingtechnology for the provision of new attractive services based on the complex composition andaggregation of virtualized resources



29/31


Conclusion

We can hence conclude that Grid Computing, the newest avtaar of distributed computing

is one of the most innovative and productive use of technology by mankind.

The future would be in terms ofCyber-infrastructure , which would include all levels of

computing, from handheld devices to supercomputing machines, and the networking and

software to enable them to work together. No wastage would be the bottom-line in the

technology sector too and maximum performance would be extracted from each and every

digital circuit working together all for the benefit of mankind.



30/31


Recommendations

With the advent of truly global organizations which have huge computing needs, Grid

computing comes as a welcome cost saver which can also be modulated to become a

profit center.

With the GLOBUS toolkit as a standard in place now, companies should focus on

harnessing the raw power of their present IT infrastructure.

Cost saving models should be grid-centric, and the implementations should end up

paying for themselves in a few years.

Major software applications such as Database software, ERP and CRM along with MIS

needs should be implemented on GRID architecture.

High speed communication links would act as a one time investment in tying all the

above mentioned facets together.



31/31


Bibliography

www.ibm.com/grid

www.oracle.com/grid

www.hp.com

www.testcases.com/gridcomp/index.htm

www.sunmicrosystems.com

www.sungrid.com

www.globus.org

www1.it1.bell.com/g/grid/impl

www.nextgrid.com

www.howthingswork.com

www.motorola.com

www.intel.com

References

Grid Computing by Peter Nash

Globus Toolkit 3.0 Documentation
http://www.ibm.com/gridhttp://www.oracle.com/gridhttp://www.hp.com/http://www.testcases.com/gridcomp/index.htmhttp://www.sunmicrosystems.com/http://www.sungrid.com/http://www.globus.org/http://www.it1.bell.com/g/grid/implhttp://www.nextgrid.com/http://www.howthingswork.com/http://www.motorola.com/http://www.intel.com/http://www.ibm.com/gridhttp://www.oracle.com/gridhttp://www.hp.com/http://www.testcases.com/gridcomp/index.htmhttp://www.sunmicrosystems.com/http://www.sungrid.com/http://www.globus.org/http://www.it1.bell.com/g/grid/implhttp://www.nextgrid.com/http://www.howthingswork.com/http://www.motorola.com/http://www.intel.com/

karan maini - roll.no. 1

Documents