Companies may have found several benefits to grid computing. But it is not yet the panacea of all problems associated with enterprises’ high computational demands. Leong Khay Mun elucidates on the technology—sans the hype

Under pressure to maximise shareholder returns, businesses are constantly on the lookout to getting more out of the company’s IT horsepower purchases, especially for the power-hungry R&D functions.

When Bristol-Myers Squibb was looking for more computing power, it had the option of paying $500,000 for a Linux cluster (where multiple servers running Linux operating systems are linked together) that might give it three times more performance than its system was capable of, reported InformationWeek.

To get a five- to ten-fold boost, the $19.4 billion New York pharmaceutical company could spend several million dollars to get a high-end symmetric multiprocessing (SMP) system (where multiple CPUs share the same memory).
But if the company could tap into the power of “Several thousand PCs” with the help of grid computing, “We can get a 100-fold performance increase,” said Rich Vissa, the executive director overseeing IT for R&D. “That was the carrot for us.”

Grid expectations
The notion of grid technology is adapted from the electrical power industry, where a lack of computing capacity in one part of the network can be compensated for by bringing in excess capacity from other parts of the network, not necessarily from the same company.

All a user has to do is to submit a calculation, the job’s CPU, and memory requirements to a network of computers linked by grid computing software, and the software will poll a directory of machines to see which has the capacity to handle the request fastest.

This resource is valuable considering the amount of computing capacity that is left idle which companies can tap into. InformationWeek reported that Intel servers only operate at about 5 percent to 20 percent of capacity during the workday, and more or less zero at night.

A company that wants to tap into a grid needs some basic components: a PC that is connected to the Internet or a secure private network, and software to search for resources available and schedule jobs, said Simon See, director, High Performance Computing (HPC) with Sun Microsystems, Asia Pacific.

On average, it costs $200 per PC for software licences, plus a couple of dedicated administrators, and the price of getting applications to run in parallel. For a 500-node grid, for instance, the price is estimated to be $250,000.
According to See, there are basically three types of grid that a company can tap into:

• Cluster grid: Connects resources within a department, like a research lab,
• Campus grid: Connects resources within an enterprise, like between departments A, B and C.
• Global grid: Connects resources from different institutions and enterprises.

Although grid computing is a relatively new concept—the term was coined less than 10 years ago, according to Intel—chances are, a company may already be deploying what the chipmaker says is a precursor of grid computing, called clustering.

Clustering refers to multiple servers that are linked together in order to handle variable workloads or to provide continued operation in the event one fails.

But because the problems companies face are getting more complex by the day, they need bigger scale clusters or supercomputers, which can get pretty expensive, said William Wu, Intel’s program manager, Asia Pacific, Itanium Processor Family.

This is where grid computing comes in.

In an ideal global grid environment, companies in the pharmaceutical, electronic design automation, energy and banking sectors can tap into far-flung computers, databases, and scientific instruments at a lower price, since they pay only for what they need.

Gritty problems
There are reasons why, in spite of its advantages, grid computing is not yet the panacea of all problems associated with enterprises’ high computational demands.

The complexity of managing the grid increases as more resources and parties are involved. Most companies like the Volvos and BMWs of the world are working within the limits of the campus or cluster grid, and not the global grid, because of security issues, said See (See box: Grid computing: the flip side).

And in a global grid, having a common standard where different systems can communicate with one another is important. “Standards for grid computing are still evolving,” said Wu, referring to the Globus Toolkit that has emerged as the de facto standard for grid computing. This working or alpha version of the toolkit is expected to be finalised by the end of 2002.

Besides, with the hype surrounding grid computing, people have formed wrong expectations of what it can do, and do not understand what it should be used for.

“A lot of people think that since I have a lot of resources available to me through the grid...I can then throw one application into the grid and expect it to run faster. But this is just hype,” said See.

What grid computing is good for is statistical analysis and scenario simulation, where you submit the algorithmic rule and the grid will help you with the calculations. A bank, for instance, may use grid computing for risk analysis computation, said See. In this instance, the bank will send out 100 scenarios at the same time for the grid to compute the results, instead of having to send scenarios one by one.

Grid computing is not for business-computing tasks such as managing the flow of raw materials and finished goods in a supply chain or selling products through an e-commerce website.

So essentially, one of the main business benefits of grid computing is that it increases worker productivity by maximising the resources you own or which are available to you.

HP, for instance, claims that with its version of grid computing solution, the HP Utility Data Centre (HP-UDC), the total cost of ownership (TCO) of a data centre can be reduced by up to 50 percent, since it consolidates resources for the provisioning of new services or applications, said Ross Templeton, Solution and Software Programs (UNIX) manager for Asia Pacific, HP.

But it does not mean that with grid computing, you can cut back from buying new systems. “With results coming in faster, this means engineers will want to find out more things. They will therefore submit more jobs and to do this, they need more machines,” added See.

Another problem with grid computing that is holding back commercial adoption in the Asia-Pacific is that the infrastructure for corporate use is not widely available—IT service providers have yet to provide a hosted grid for companies to tap into immediately.

To get started, See suggested companies start with a cluster grid and then expand the grid stage by stage.

— This article first appeared in Intelligent Enterprise Asia