Grid Computing

A grid is a ‘web’ scale federation of computing, networking and storage resources. The resources can contribute to the grid infrastructure capacity in a primitive sense by providing CPU cycles and storage, or they can provide grid services such as access to information repositories and applications or algorithms. Examples of information repositories are news archives of leading newspapers, geographical maps and traffic conditions. Examples of grid applications are service for translating a news item from one language to another, algorithms for forecasting price movements or algorithms for finding the optimal driving direction to a chosen destination in light of the prevailing traffic conditions.

Web vs the grid



What does the grid provide over and above the traditional Web? Versions of data repositories mentioned above are available on the Web. The websites providing access to data often provide applications that can process the data. The best way to define the grid is to contrast it with the Web. The differences listed below stem from the fact that while human beings access the Web, programs will access grid services.

Human beings are resilient to formatting styles. If we ask for the distance between two cities, irrespective of whether the answer is represented as an integer, decimal number or a fraction, we comprehend the answer. Computers, on the other hand, do not have a universal standard to express numbers or text. Hence Grid services rely on a standard for message syntax called SOAP (Simple Object Access Protocol) to encode messages and service requests. SOAP protocols use XML to tag data with semantic annotations. 

People can follow the instructions on the website written in natural language to compose the request for a service or to search for information. Programs that access web services on the Grid use predefined access methods expressed in WSDL standard, again based on XML. The service description includes the information contained in requests for service and their responses.

The analog of search engines on the Web will be service registries based on UDDI specification. The UDDI registry of a service will contain the URL and the description of the service in WSDL, the latter incorporating description of the SOAP requests it accepts and the responses it generates. Applications on the grid thus have the ability to automatically access information from multiple servers at diverse locations on the Web and to process this information using multiple application providers also at diverse locations on the grid.

The most remarkable feature of Grid absent from today’s Web is the ability to compose higher-level services from basic services and information repositories without human/programmer involvement or awareness of the fact that geographically distributed resources are being accessed. The orchestration of accessing data and moving it to the applications that need it, and the movement of results produced by these applications to other applications when needed, is defined in a standard language BPEL (Business Process Execution Language) developed for expressing workflows for Grid Services. 

Application outsourcing 



Businesses will change their model for managing apps. Instead of maintaining all applications in-house, apps that are essential but not strategic differentiators will be outsourced. Responsibility of ensuring that the application stays up, managing upgrades of the operating system and middleware of the application server and fixing the occasional bug will become the grid service provider’s responsibility. Furthermore, the application will be outsourced generically to the multitude of vendors who provide desired service with identical programming interfaces. 

Resource Grid: Grids can also be the infrastructure providing CPU cycles and storage over the Internet. FTP solves the

problem of sharing files over heterogeneous file systems and storage architectures. The Java family of standards like J2EE will

allow applications to migrate from one compliant platform on the Grid to another to make use of available processing power.

However, to realize the promise of Grid, several technologies still need to be developed, some of the key ones being:

As resource owners place their resources on the Grid, they need the ability to define the rules or policies for sharing the resource, along with software capability to enforce these policies/rules.

To enforce the above policies, the usage of the resources has to be monitored 

Security mechanisms will be needed to guard against unauthorized use of the resources. Grids services may be accessible to anyone for free or a fee, or their use may be limited to a business or a business and its business partners. 

At the Grid level, availability of resources and the load on them will have to be monitored

The above requirements are also essential for service grids. The responsibility of finding the application automatically from the available ones and assigning the

service request to it has to be managed by Grid software. 

The benefits of sharing CPU cycles over the Grid are most obvious for big science engineering projects, which at times need substantial resources for computation. It is unfeasible to provide for the maximum anticipated computing need of each project and it makes sense to be able to share the computing resources among projects. Data centers of the seventies attempted to address the same scarcity of resources, but the rules of sharing the resources were too rigid for day-to-day operations and sharing was possible only in very tightly knit organizations. Grid technologies will allow sharing to happen across loosely knit organizations.

Some examples



Currently, most Grid deployments are in the US and Europe. However, the potential for grids in India and other developing nations is significant. A large number of Indian academic and research institutions are active in the area of Biotechnology. Collective investment by them to create a Grid to pool their computational resources and to share their scientific data will benefit the entire bioinformatics/biotechnology research community. Another good example is the e-Governance. Some examples of Grid deployments in the West are:

The University of Pennsylvania Grid will enable thousands of hospitals to store patient mammograms in digital form so that patient’s medical records can be accessed collectively by his physician or authorized researches. The Grid also makes analytical tools developed by various researchers to the medical community to help diagnose individual cancer cases and to identify cancer clusters in the population. The grid has three tier architecture consisting of a temporary repository at each hospital, metropolitan hubs and a regional hubs for the mammogram data. The latter two are implemented using Linux workstation clusters.

The North Carolina Biogrid is resource grid as well as a services grid. Three Linux clusters and a top of the line RISC workstation are the computing resources shared by educational, research and commercial applications. Duke University, University of North Carolina, North Carolina State University and North Carolina Supercomputing center are among the participating organizations. 

UK computing and data grid will use the utility model of computing to offer computing resources to the scientific community. The resources will be located at a national center and eight regional centers. Part of the grid is a shared data facility at Oxford University to give the scientific community access to experimental data from high energy physics experiments conducted around the world.

Just as the HTTP protocol and HTML standards made it possible share content published anywhere in the world to the international community, the standards for defining computational resources and data/application services, along with relevant software will allow owners of computational/ storage resources and information/application services to create a rich array of service Grids.

Dr Manoj Kumar, Director, IBM India Research Laboratory