Gotcha! Don’t Let Grid Computing Trip You

Grid computer systems—clusters of many cheap computers—provide an economical alternative to supercomputers, particularly for scientific and engineering problems like simulating an oil field’s yield. But there are obstacles for deploying these setups: They pose basic computer science challenges, standards are unsettled, and software licensing schemes are only beginning to adapt.

PROBLEM: Applications only derive benefits from grid computing if they can achieve an efficient division of labor.

RESOLUTION: Muster all the distributed and parallel computing expertise you can in advance. The trick, says Colin Wheeler, a systems consultant who worked on a grid system deployed at the Royal Bank of Canada, is in determining how you can cut up the actual piece of work that you need to do. “First ask yourself, is this actually a calculation that I can farm out into different components?” he says.

PROBLEM: Commercial software may not be practical in a grid environment for technical or licensing reasons.

RESOLUTION: Don’t assume you will be able to run your packaged software in a grid environment. Even if software is technically ready for grid deployment, traditional per-CPU licensing schemes may get in the way.

For example, instead of running a supply chain planning analysis on a high-end 16-processor server, in a grid environment you might want to farm out the job to 160 single-processor servers that can accomplish the same job in a fraction of the time. However, that might not be a good deal if the vendor expects you to pay 10 times as much, even though the software was only loaded and running on those machines for a short period of time.

William Fellows, an analyst with the 451 Group, says vendors are starting to take “baby steps” toward adapting their licenses to the realities of grid computing.

PROBLEM: Bandwidth constraints and latency (the delay in data transmission over a network) can undermine the efficiency of a grid-based system.

RESOLUTION: Try to select applications that require a minimal amount of data transmission between nodes. While a grid of computers communicating over Internet protocols can equal or surpass the computing power of a high-end machine, data transmission over a network is slow compared to data transmission between processors in the same computer cabinet. This is a factor for geographically distributed grids, as opposed to clusters of computers within a data center.

Transmitting large amounts of data also eats network capacity and costs money, so that the cost of distributing work to the nodes in a grid could outweigh the savings from using less expensive computers.

Story Guide:
Straight shooter: CEO Rex Tillerson Doesn’t Play Games

  • Big Oil’s Big Challenge
  • R&D Like Launching a ‘Moon Shot’ Every Decade
  • Exxon’s Balanced Methodology
  • Virtual Drilling: Exxon’s Technology Edge
  • Business Intelligence Guides Exploitation of Oil Resources
  • Oil in the DNA: The Exxon Culture
  • ExxonMobil Base Case
  • The Technology of Oil Exploration

    Context:

  • How Big Oil Uses XML
  • Roadblock: Getting Oil Business Managers on Board
  • Gotcha: Don’t Let Grid Computing Trip You
  • Vendor Profile: A Look at IBM’s Supercomputers and the Oil Business