Big iron vendors bet on grid computing to meet explosive demand for compute resources MORE POWER. It’s a concept that appeals to just about every technology enthusiast. Thanks to continuous advances, we hold fast to the promise of faster processing, more storage, and unlimited information access from any device. And so we turn to grid computing. Similar in concept to power plant grids that channel electricity where it’s needed, grid computing aims to deliver data processing resources including CPUs, storage, and applications across nations and continents. We’re talking about serious computing here. The scientific community and organizations such as CERN (the European Organization for Nuclear Research, which gave birth to the Internet) are working on ways of processing literally petabytes of data. Standards bodies such as the (Object Management Group (OMG), World Wide Web Consortium (W3C), and Grid Forum are already coalescing around the idea that formal guidelines are needed to build a vast distributed architecture that leverages technology like XML and peer-to-peer networking. And supercomputing vendors such as IBM, Sun, Hewlett-Packard, Compaq, SGI, and Cray have woken up to the idea. IBM started its Grid Computing Initiative last year, and Sun recently announced its Technical Compute Portal offering, comprised of the iPlanet Portal, Sun’s Grid Engine, and Sun ONE (Open Net Environment). The appeal of grid computing to these vendors goes beyond the opportunity to ship big iron. The big question still floating around the scientific community is, “How much processing power is enough?” A popular line of thought suggests Moore’s Law cannot keep pace with the need of large enterprises for more CPU cycles, unlimited storage, and boundlessly distributed applications. Of course, the industry attempted to address this problem with what used to be called distributed computing in the 1980s. But if you take a look at a few examples, you quickly get the picture that this is very much uncharted territory. Current projects are centered on research in fields such as earthquakes, the ozone layer, particle accelerators, and human genes, each collecting mounds of data that scientists in other parts of the world may need for related projects. Where business is concerned, real working examples are hard to find. But it’s not hard to imagine that in the future it will become increasingly difficult for existing supercomputers to analyze data contained in vast databases spread across many geographical areas. Some of the challenges ahead require solving numerous problems such as security, connectivity, and semantic differences. For example, extracting data from heterogeneous databases requires having global directories and data dictionaries helping to locate and select only the pertinent information. Technologies such as UDDI (Universal Description, Discovery, and Integration) and XML can help, but if grids are to be successful they must be easy to use and devoid of ad hoc programming. Interestingly enough, Oracle told us last week they want to develop “the database for the grid.” Today, Oracle’s biggest databases handle between 30TB and 40TB of data, but executives are predicting huge growth rates. If it all sounds too far off, start thinking on a smaller scale. Grid computing can become the logical evolution (and consolidation) of multiprocessor servers and computer clusters, tied by technologies such as Infiniband. Take a second look and you will see an architecture similar to a computing grid that can flexibly allocate CPU cycles and terabytes to your applications. The additional power and flexibility that enterprises need are there, but we have to learn how to exploit them before the future catches up with us.