Belkin • Compton, Calif . • "Are you crazy?" That was the reaction John Adcock, network services manager, got when he told the vice president of finance that he planned to port applications from 11 Sun data center servers to one 64-processor HP 9000 Superdome.
After all, the servers were running a highly customized version of Oracle's 11i database and enterprise applications . The consolidation would require a complete data conversion, including 241 patches, some of which can take hours to apply. And this wasn't the only IT project the fast-growing interconnectivity manufacturer had going on at the time.
But there was a method to Adcock's madness. His Sun Solaris 6500 database server not only needed upgrading, but it also had crashed six to eight times during its two years in service. Adcock wasn't so sure he wanted to stick with Sun, but he also didn't want a mixed environment. One day, while researching server prices, the Sun server crashed for the last time. "I realized that I had the ammunition to change this," Adcock says.
That's when the serious homework began. It took two months to assemble the data needed to compare the three-year costs of purchasing, upgrading, maintaining and supporting several Sun servers vs. purchasing one high-end HP server, converting the data and training his staff on HP-UX. Other considerations included floor space, power consumption, administration, manageability, labor costs and connectivity costs.
Consolidation turned out to be the less-expensive option. Even so, "I had two to three weeks of sleepless nights after making the decision,"Adcock says. "It's one of those decisions I knew would make or break a career."
The consolidation has had many benefits, including a move toward utility computing . For instance, Adcock bought a 64-processor Superdome, but only paid for 48. The unused processors are priced at "instant capacity on demand" prices and can be "turned on" and paid for as needed.
Adcock says he also likes the enhanced fault-tolerance capabilities afforded by partitioning. "I have nine instances running on the Superdome; technically it functions like nine separate servers," he says. As a result, if a network card goes down or an app misbehaves, none of the other applications are affected. If a processor fails, an auxiliary processor kicks in.
But as Adcock anticipated, the move wasn't easy. His team had almost finished when it got word that it had one month to upgrade its Oracle applications to support the company splitting into two separate entities. Even though this entailed halting and then starting over again on the conversion, it actually made it easier because it eliminated the need to transfer patches from the old version of the application. "We had a team of about 10 people do it in a week," Adcock says.
It all ended well, but Adcock says the main lesson he learned was that it's important to know all the major projects planned across the company for the next six months. Other advice he offers: