Low-hanging fruit? Fuhgeddaboudit. For its grid foray, UPS chose its all-important Flexible Bill Rendering application, which generates invoices for customers in the United States and 35 other countries. The volume is staggering - UPS produces 4 million to 6 million invoice pages weekly for mailing and posting online, says Brian Cucci, manager for UPS' Advanced Technology Group in Mahwah, N.J.

As a mainframe process, the U.S. portion of the invoice-composition application took about 20 hours per weekend. The grid requires a fraction of the time the mainframe needs to do this work, and not just because the Linux boxes are faster. In a test on a proof-of-concept grid built in 2005 (using spare CPU cycles on the IT staff's laptops), an analytics application took 45 minutes to complete what was taking stand-alone servers 9.5 hours, Cucci says. Encouraged by that finding, UPS built separate grids for its two primary data centers. Built with DataSynapse's GridServer, each grid is composed of four two-way commodity servers that run the grid engines; other servers manage the grid, he says. Each grid has the primary responsibility of running different applications - they now host three applications. The second grid acts as business continuity backup for the bill-rendering application.

--Julie Bort

< Return to main story: How to avoid bumps on the road to grid computing >