Making the case for the new data center

Last June, John Weeks stood in the data center at the Riverside, Calif., headquarters of law firm Best Best & Krieger envisioning a forklift. As the newly hired IT director for this six-office, 310-employee firm, he had to solve some pressing problems inhibiting aggressive growth plans.

The 100-year-old law firm, one of California's largest, had evolved without a formal IT agenda despite relying on mission-critical applications for functions such as document management and billing. Weighing the firm down were an aged Novell NetWare 4 network, a patchwork of desktop operating systems, ancient e-mail and word processing platforms, plus inadequate security, bandwidth and systems management.

Weeks set about transforming the rickety IT infrastructure into a model of the new data center - a feat he wanted done in six months.

Experience with the old, constantly failing IT systems made BB&K partners and other employees wary of trusting a fully centralized data center, Weeks says, so he wanted a design that would allow offices to function separately. Plus some of the firm's critical custom billing applications wouldn't "play well with other applications" when sharing hardware, he describes. A traditional design would have placed these on their own servers and included separate mini data centers at each site - an expensive approach that would waste a lot of hardware capacity. Virtualization, a tenet of the new data center, provided the answer.

Integrator Agile360 pitched a data center design that featured virtualization software from VMware (recently acquired by EMC). With virtualization, even those anti-social applications could be made to share servers. The virtualization product "isolates each instance of an application, so the application doesn't necessarily need its own hardware," Weeks says.

Moreover, virtualization gave Weeks the cost-efficient redundancy he needed, as "virtualized" primary servers can be backups, too. By encapsulating a specific virtual machine (an application and its operating system needs), any application can be nearly instantly ported to any available server.

Weeks bought more RAM for two HP ProLiant DL380 servers he had recently installed at the main Riverside data center (for 8G bytes of RAM) and loaded VMware's ESX Server software on them. He similarly upgraded and outfitted five ProLiant ML370s installed at the remote offices. This gave the data center many machines that could virtually operate as one - or as redundant servers. Weeks consolidated 54 outdated servers into 16 new servers running the VMware software.

With the addition of Microsoft Clustering Services, which Weeks is implementing now, the data center servers also can fail over to one another while offering guaranteed performance even for demanding e-mail and database applications. They also scale easily.

Because clustering requires a storage-area network, Weeks built a hefty 100G byte Compaq 10000 Fibre Channel SAN in the Riverside data center. "My design goals were to not have to buy hardware for three years," Weeks says. Compaq Insight Management server management software gave the data center self-management functions such as auto discovery.