Oracle CEO Larry Ellison's $100-million baby - storage start-up Pillar Data Systems - is expected to deliver as early as June what analysts are calling a unique storage subsystem that stores data on disk based on its importance to a customer's business.

The company hasn't been talking - and still isn't - but a wealth of information recently posted to www.pillardata.com paints a fairly detailed picture of what the company is planning.

Founded four years ago in San Jose, Pillar Data is funded by Ellison's private investment group, Tako Ventures, to the tune of a reported $100 million. Lawrence Investments, the parent company of Tako Ventures, has interests in approximately 30 technology and biotechnology companies. Pillar Data employs about 300 people and is headed by Mike Workman, a veteran of IBM and Connor Peripherals.

Pillar's Axiom Storage System is expected to let customers combine the storage resources from network-attached storage (NAS) and storage-area network (SAN) environments into a single pool that can be managed and accessed from a single console.

Business-critical database or transaction-oriented data is stored close to the outside of the disk, where its seek time is less and where it can be quickly accessed. (Seek time refers to the amount of time it takes to position disk heads so data can be read.) Less-important data such as e-mail is stored closer to the disk spindle, where access takes longer. The least-important data - file and print data - is stored on the inside of the disk where the seek time is longest.

Rules-based software called the Pillar Storage Manager would let IT administrators create policies that determine the placement of data across these three tiers of storage.

This concept of storing data by disk spindle is not new but is as yet untried in open systems - Unix, Linux and Windows storage. IBM last tried it with its Direct Access Storage Device mainframe storage.

"It is a very uncommon approach," says Randy Kerns, senior analyst for Evaluator Group. "Years ago, we tried to do things like that with IBM mainframe disks and it was really too much to manage. Doing this in an open or distributed systems approach is unique. There's a lot of skepticism whether that's a good idea or not."