Grid Computing—it’s simple, but is it easy?

Grid computing provides an adaptive software infrastructure that makes efficient use of low-cost servers and modular storage, says Arunava Dutta

Over the last couple of years, companies have attempted to make efficient use of their IT resources. To do this, they have to deal with islands of computing systems that have sprung up within their organisations, and that are slow to change and expensive to maintain. Research reveals that many companies simply buy more—or more powerful—hardware when key servers get near capacity, rather than seek ways of using current resources effectively. (Source: ‘Beyond Infrastructure’, QNB Intelligence, October 2003.) This builds up the loaded islands but leaves much capacity underused in other islands.

Grid computing has caused such a buzz because it actively addresses this key issue.

It provides an adaptive software infrastructure that makes efficient use of low-cost servers and modular storage which balance workloads more effectively and provide capacity on demand.

Scaling small servers in small increments provides flexibility, performance and reliability at low-cost. According to Gartner, “Enterprises evaluating high-performance computing systems should investigate the potential fit of grid technology as a means of lowering the cost of equally powerful traditional equipment or raising the effective performance to reap additional benefits.” (Source: ‘Grid Technology Is Influencing the Future of Large Servers’, C Claunch, June 10, 2003.)

New unified management allows you to manage everything cheaply, and in a simple manner, in the grid. Giga Research has stated that “Grid computing can provide cost savings as an alternative to building out a company’s own hardware, storage and networking infrastructure.” (Source: ‘The Next Big Thing: Grid Computing’, Stacey Quandt, March 4, 2002.)

Grid computing is said to be a “disruptive” technology, just as the Internet was—it fundamentally changes the way things are done. The early Internet days saw an outburst of hype that the IT industry is only now beginning to recover from. But the fundamental truth is that the Internet has changed everything, just as was promised—it just took longer than expected.

Similarly, there is a lot of hype about grid computing, as is apparent from the number of different names it has been given as vendors try to differentiate their vision from everyone else’s: Computing on Demand, Adaptive Computing, N1, Utility Computing, Hosted Computing, Organic Computing, and Ubiquitous Computing are just some examples.

There is, though, a problem. It is this—the grid proposition is simple, but making it happen is not necessarily easy. It all depends on the approach. Says IDC’s Dan Kusletsky, “Some people are frightened of grid computing. They think it’s going to be complex, they think it’s going to be difficult, and if you’re fully implementing a thousand-node cluster doing computational work for the weather service, it can really be very difficult. But that’s not where you have to start. You can start by taking the applications you have today and put distributed database functions underneath it, and make the database, the grid, your database engine.”

Coming to grips with reality

Despite this, don’t think that grid computing is a myth. It’s not. Companies everywhere are beginning to reap benefits from the use of grid computing technology. They are doing this not by adopting “utility computing” solutions built on the vision of a scientific grid like the SETI@home project, but by taking things one step at a time, from the inside out. (The Search for Extraterrestrial Intelligence —SETI—is one of the earliest examples of a scientific grid. Signals from telescopes, radio receivers and other sources monitoring deep space are distributed to the PCs of individual science buffs via the Internet. This loose network of small computers crunches numbers, looking for patterns that could suggest signs of intelligent life.)

The SETI@home principle of harnessing idle computers across the Internet is intellectually interesting, but businesses will never want their data or their computing resources distributed to random computers. Still, just as businesses have introduced the concept of the public Internet in-house to make intranets, companies can bring the concepts of the scientific grid to make enterprise grids. And they can start—and some have already started—doing that.

For example, Electronic Arts has launched an online version of its popular PC game, The Sims. The Sims Online is hosted on a grid built on commodity clusters providing and managing the capacity needed to support online gaming by 100,000- 150,000 concurrent users. CERN is exploiting low-cost commodity hardware to build a grid consisting of thousands of such machines to handle the data produced by its Large Hadron Collider project. Oracle uses a grid in product development—hundreds of its developers’ machines that are unattended at night are transformed into a grid to perform regression testing on the latest code developments.

Enterprise grid computing lowers costs by:

• Increasing hardware utilisation and resource sharing.
• Enabling companies to scale incrementally with low-cost components.
• Reducing management and administration requirements.
• Building critical software infrastructure that can run on large numbers of small, networked computers by combining two related concepts:

1. Implement One from Many. Grid computing coordinates the use of a cluster of machines to create a single logical entity such as a database or an application server. By distributing work across many servers, grid computing exhibits benefits of availability, scalability, and performance using low-cost components, as well as the flexibility to respond to rapidly-changing business requirements. Because a single logical entity is implemented across many machines, companies can add or remove capacity in small increments online. With the capability to add capacity on demand to a particular function, companies get more flexibility for adapting to peak loads, thus achieving better hardware utilisation and better business responsiveness.

2. Manage Many as One. Grid computing allows you to manage and administer groups of machines, groups of database instances, and groups of application servers at low-cost. Grid computing first removes many of the administrative costs of managing a single system by making each database and each application server adaptive to changing circumstances. The model then makes handling many systems simple by allowing them to be managed as a single logical entity.

The requirements of a grid computing infrastructure can be described by the following attributes:

• Virtualisation—the abstraction into a service of every physical and logical entity—at every layer of the computing stack.
• Provisioning of work and resources—distributing supplies where they are needed based on policies and dynamic requirements.
• Pooling of resources to increase utilisation.
• Self-adaptive software that largely tunes and fixes itself.
• Unified management and provisioning.

The journey to grid computing

The irony of this is that, all hype aside, many businesses are already taking the right steps towards grid computing, simply as a part of getting their IT infrastructure under control. Many will choose to undertake this migration by starting with small-scale pilots. As more and more companies deploy clusters of industry- standard servers, IT infrastructure resembling enterprise grids will naturally result. The trick is to do things in the right order. We see three steps that companies should take on their journey to grid computing:

Standardisation on low-cost, high-density modular servers and storage based on technology such as Intel Itanium processors, blade servers, and Linux or Windows.

Consolidation of clusters of servers and storage shared among one or more data centres.

Automation of all day-to-day management tasks, enabling a single administrator to simultaneously handle hundreds of servers in clusters.

Each stage brings its own possibilities for savings and efficiencies. Gartner estimates that typical enterprises with mainframe, Unix and Windows deployments could save between 8.5 and 10.5 percent of the data centre budget by implementing standardisation, consolidation and automation (Source: ‘The Impact of RTI on IT Operations Budgets’; Donna Scott, John Oborn, Barbara Gomolski; Gartner Research Note; July 17, 2003.) And according to Giga, companies can potentially save 20 percent or more through consolidation. (Source: Giga Group, May 2003.)

Many vendors are leaping on to the grid computing bandwagon and adding to the hype surrounding it. But companies must realise that moving to grid computing is a journey, and probably not a short one. Over the next few months and years, more companies will join those who are already several steps along the way to realising the benefits that will accrue from grid computing.

The author is Director, Technology, Oracle India.
E-mail:arunava.dutta@oracle.com