May 28, 2005

AS THE data to be analysed increase exponentially, the analytical power required to process applications also shoots up. Put differently, the needs of computing are increasing fast. It is important, therefore, that the computing world develops capabilities fast enough for it to meet the rising needs.

This is the reason why the computing world longs for a technological breakthrough. The solution may lie in the concept of grid computing.

Grid computing is the concept of establishing a community where resources are offered for use under some organized and controlled system. Grid computing architecture is meant to complement, rather than to compete with, the existing networks. It can provide dynamic access to a large pool of resources.

Virtual Organizations (VO), developed in grid, could dramatically change the way computing is done, just as the internet has revolutionized how information is shared. Anyone with excess resources can join and thus add to the power of a particular grid and get paid in the process.

This way the computing world may cope, in a cost-effective manner, with the requirements of organizations having ever increasing data and varying processing speed requirements. This doesn’t, however, imply that grid computing has made traditional high performance computers obsolete. Rather, it has served to increase their demand by making access to them easier.

The concept

The concept of grid computing is similar to that of an electric power grid, where consumers can “plug in” to get the required power. A large source of computing power can be established simply by combining many available sources.

Grid computing is not confined to CPU cycles. A grid can be thought of as a computing resource, also referred to as a service, like CPU time, storage, data, code repositories and software.

The challenge is to establish an actual grid composed of less-expensive computing resources in an effort to generate greater power for every rupee invested. This way a grid may perform better as compared to expensive, high performance sets of machines.

The common problems faced in the setting up of a global grid include user authentication, resource authorization and access, discovery of resources and sharing, and security of assets on the consumer and provider ends.

Grid architecture is based on components that interact with one another to establish a VO. This component base is similar to the architecture used in other problem-solving disciplines of software development.

There are different issues that need to be addressed at this level. Interoperability is a fundamental concern, which involves establishing relationships among different parties and across different platforms.

Services are important because they represent the ultimate requirements of the grid. Services that can be requested include data, CPU time, storage, network resource and software.

Application Programming Interfaces (APIs) and Software Development Kits (SDKs) are also required so that grid aware applications could be developed. Grid architecture is designed in layers to address all these issues.

The layering type known as “fabric layer” deals with interfaces to provide local control of resources. Typical interfaces are required for monitoring and controlling the execution of processes and their enquiry functions to determine the state and load.

Network resources are needed for prioritization and reservation. Catalogue resources include databases, whereas code repositories and other storage resources require interfaces for their specific resource information and enquiry mechanism. Grids use existing protocols and interfaces on fabric and/or add other interfaces where required.

General Purpose Architecture for Reservation and Allocation acts as a “slot-manager” for missing functionalities. Condor is another implementation plan for advanced reservation support.

As opposed to the “fabric layer”, the “connectivity layer” provides for communication more easily and securely. Communication includes transporting, routing and naming functionalities.

‘Single Sign On’ is an already in use mechanism to save from repeated user intervention. Connectivity on grid also involves working with local security mechanisms like Kerberos security. User based trust relationship also needs to be established between sites for connectivity. This grid architecture uses existing internet protocols for security with modifications and/or enhancements like Grid Security Infrastructure (GSI).

The “resource layer” deals with sharing of single resource functions. This includes APIs and SDKs for secure negotiation, initiation, monitoring, control, accounting and payment of sharing operations involving single resources. Information and management protocols are required to obtain information about the structure and state of the resource and then to negotiate and establish sharing relationships.

Grid Resource Information Protocol (GRIP) is based on LDAP (Lightweight Directory Access Protocol) to define standard resource information. Grid Resource Registration Protocol (GRRP) is an associated protocol. There are also C and Java APIs and SDKs available for client and server side integration of resources into the grid.

The “collective layer”, as a global and coordinating layer, works to capture interactions across collections of resources. It is a heavy functionality layer that requires a number of services.

Directory service is required to discover the existence and properties of VO resources. Brokering and scheduling services are needed to request specific purpose services and scheduling of tasks.

Community authorization servers are required to enforce community policies to govern resource access. Community accounting and payment services are meant to handle accounting and payments mechanisms. Specific implementations are available to provide different services at this layer, like DRM broker, Nimrod-G, Akenti, CAVERNsoft and Access Grid.

The perspective

The grid is thought of as the next generation internet. Actually, it is not an alternative to the internet but represents a set of additional services that build on its protocols and services. As a result, resources on the grid are also on the internet, by definition.

The grid provides a source of free cycles, besides other computing features. Typically, resource owners will enforce policies to limit access based on membership or accounting information. Hence, the architecture must incorporate the ability to calculate individual and collective resource exchange and cost information.

The grid requires programming models to deal with problems not encountered in sequential or parallel computer systems. However, fundamental concepts like abstraction and encapsulation in programming models can reduce complexity and improve reliability in application development.

Millions of processors accessible within a VO represent a significant source of computational power, comparable to or even higher than high performance computers. However, many problems require tightly coupled computers and grid computing increases, instead of reducing, the demand of the machines.

Challenges and future directions

As the concept is still evolving and is rather immature, grid computing poses a number of problems. The security of computing assets at the consumer and provider ends is critical, for instance.

Grid applications involve sharing of resources including processor, memory, code and data. At the consumer end, data, processing application and integrity of results may be at risk. Provider end securities involve agreed use of shared resources between distrustful parties and security of other unshared assets. Accounting, calculation, payment and resource costing issues also require transparent and guaranteed solutions. A grid of distrustful parties requires conflict resolution approaches with the architectural support of documented and trusted evidence mechanism.

Future development of grid technologies, intended for internet environments, will involve the development of sophisticated grid services and the evolution of services for the consumer market. Among others, Commodity Technologies, Legion and Globus are expected to name architectures that have been fashioned with different interest groups in mind.

The Globus approach, for instance, is a result of the community-based initiative Global Grid Forum. The aim is to support the development, implementation and deployment of grid standards and application development.

The initiative is based on the assumption that grid architectures should provide basic services only and should not prescribe higher-level programming models and architectures. It has already provided a toolkit with services for security, communication, resource sharing, location and allocation.

Process management and data access services are also provided. These services are provided at a low level. Programming models and tools can be used to implement higher-level services.

Other groups are also active. All these efforts have laid the groundwork for the establishment of platform-neutral and interoperable Virtual Organizations, thus enabling grid.

The writer is a student at the Muhammad Ali Jinnah University

Please Visit our Sponsor (Ads open in separate window)