A start-up run by the former CEO of Cobalt Networks is readying a multi-core server designed to speed up Java processing in the data center.

Azul Systems, a 2-year-old, 130-person company, expects to begin selling the server in the first half of 2005, the company said. The company's president and CEO is Stephen DeWitt, who served as vice president and general manager of Sun's server appliance division after Sun purchased Cobalt (a Linux systems vendor) in 2000. Dewitt left Sun two years after the Cobalt acquisition.

DeWitt's company has developed a server appliance that works with existing Java 2 Platform Enterprise Edition (J2EE) software without modification to increase processing performance, says Shyam Pillalamarri, co-founder and vice president of software engineering at Azul.

"We have figured out a way to mount compute power remotely," he says.

Azul wants to emulate the success that Network Appliance had building storage appliances with the Network File System protocol, but Azul's appliance will support the J2EE standard used by application server software from companies such as IBM, BEA Systems and Oracle, Pillalamarri says.

Azul's server appliance eventually could be used to speed up .Net applications by supporting Microsoft's Common Language Runtime, Pillalamarri says. "We could support that in exactly the same fashion," he says. "That's not something that we're targeting right now because most of the market is J2EE."

To use Azul's product, customers must install proxy software on their servers, which then offloads J2EE processing on to the Azul server appliance. The Azul server, which has yet to be named, will be based on a custom 24-core processor designed by Azul and manufactured by Taiwan Semiconductor Manufacturing, Pillalamarri says.

The proxy software can be installed on the Windows, Linux, Solaris, HP-UX and AIX operating systems, he says.

Azul has not yet determined the exact configuration of the appliance, but it is considering a size as large as 11U with 16 processors and 256G bytes of memory. Such a system would have 384 processor cores and would be able to run Java applications, split into a large number of discrete tasks called threads, much faster than today's servers, Pillalamarri says.