Taking a serious look at grids

转载原文地址：http://www.infoworld.com/article/04/01/16/03FEgridsnow_1.html

By  Tom Yager 
January 16, 2004

Whatever future there is for business applications of grid computing, it won’t be shaped by technology. Business attitudes must change to embrace the technology that already exists. This is a classic example of technology that’s crucial to business having its uptake delayed by an apparent lack of demand. It isn’t a solution looking for a problem; there are plenty of existing problems to go after with grids and similar approaches to distributed computing. What sticks in the minds of many IT professionals is that grids are useful to science and academia, therefore they are likely to be too esoteric for business computing.

We’ve been here before. You may recall that Unix was regarded as an operating system suited to academic, scientific, and technical applications. When Unix servers were brought into business, they had to be looked after by a team of Unix Wizards. That term, along with the job it described, disappeared with the development of better management tools and techniques. We are now well beyond the days when every Unix user had to be adept at C and Perl programming. Remember, too, that it took many years for Windows server operating systems to shake the stigma of Microsoft’s involvement in home and desktop computing. Unix and Windows NT spent years stuck in their respective boxes, and I often ponder what mainstream computing lost during those years.

When businesses let grids out of their box, they’ll find that a grid is capable of tackling many of the tasks for which we now use clustering and load balancing. Grids’ most significant catch is this: To work at top efficiency, each grid node must spend more time computing than communicating. If a grid task has to carry a lot of data with it, the task won’t move easily from node to node. And if grid tasks wait in line for some precious resource (like a table-locking database), the grid offers no advantage over more common techniques.

How does IT start moving toward grids? The process of adaptation is already under way, even though the work is not obviously being done for that purpose. Business process management and related technologies are most effective when implemented closely to grids’ stateless, decentralized model. Fast systems on moderately fast networks thrive on compute-intensive units of work, along with the small amount of data that’s needed to accomplish the task. Even for organizations that decide not to implement a grid, a grid-friendly architecture is a smart way to maximize existing and inexpensive resources.