Cassatt Corporation, the glitterati start-up headed by BEA co-founder and ex-CEO Bill Coleman, is a bit sheepish about phrasing it quite like this, but the company thinks it's about to unleash the "next wave of computing."

Its problem is giving this next wave of computing a name.

There's no clear category or snug harbor Cassatt can tuck its products in just yet to make its positioning easier.

Its aim is to calm the "perfect storm" that commoditization has wrought in the data center and manage the chaotic pile of scaled-out commodity servers brought in to replace the data center's once vaunted proprietary systems.

Cassatt thinks it can do what IBM's utility computing and on-demand initiatives, as well as HP's Adaptive Enterprise, have promised to do, but haven't done irrespective of the underlying hardware, and certainly haven't done without adding complexity to complexity.

For lack of a better expression "agile enterprise" probably best describes what Cassatt is trying to do.

It thinks that since we've been dicking around with computers for 50-odd years, we might like to try IT automation for a change and let the beasts do what they're good at.

See, people aren't particularly good at managing machines and the proliferation of industry standard boxes in the data center has played havoc with IT budgets. According to IDC, 15 years ago labor accounted for 25% of the IT budget. It has now escalated to 70% or 80%.

The amount of money being spent on administration has eroded any savings the enterprise has realized by moving to plain vanilla hardware.

Cassatt claims it can solve the TCO crisis by letting computers understand and fulfill service level agreements. They're better at it than people.

In so doing it says it can solve a problem endemic to scale-out. To wit, things that work fine on one system break, and can only be mended manually with that computer version of duct tape known as scripting.

Cassatt also claims it can cure the tendency to provision for the hundred-year flood.

It says overbuilding in and of itself threatens performance and reliability.

Its cost-effective solution is to create a virtual resource pool that can handle large complex dynamic workloads by allocating processor, storage and network resources as required.

It says it can do this without disrupting anything. Its widgetry doesn't muck with the hardware, operating systems, applications, systems management or network.

Cassatt calls its idea of dynamic non-intrusive flexible scaling its simply on-demand vision.

Cassatt has been operating in stealth mode since Colman took over a year ago. It is now about to drop a few of those veils and say that Collage 2.0, the first iteration of its widgetry, has been quietly in production at a number of chi-chi sites for a while now and that Collage 3.0 should be out this winter.

Of course, Cassatt being a California company, and California being, ahem, unseasonable, it may be best to think of the timetable as being Q1 - about six months later than Cassatt once envisioned.

Collage 2.0 is a Linux-only platform; Collage 3.0, now in the proof- of-concept stage, manages both Linux and Windows 2000/2003 machines. There is a Collage 4.0 - scheduled for the second half of next year - that is meant to be broader still, but is reportedly still being defined, according to the guy defining it, Cassatt co-founder and VP, product marketing David McAllister.

Collage 2.0, which has been installed by third parties, can only manage a single application at a time. It uses virtual imaging to give a hodgepodge of scaled-out systems a single view.

Collage 3.0, which is going to be handled more directly, is supposed to manage multiple applications, giving Cassatt an entry point to the kind of software provisioning that will eventually be needed to handle tens of thousands of web services and service-oriented architecture (SOA).

Cassatt is going to focus on Collage 3.0, but won't discontinue Collage 2.0.

A 50-node Collage 2.0 implementation goes for $100,000-$150,000. Collage 3.0 is still being priced and about the time Cassatt figures out what it's going to cost, it may be more amenable to explaining how exactly Collage works.

Cassatt says it will be able to manage hundreds of thousands of processors. It says it doesn't give a hoot how the hardware is configured. It could be blades, two-ways, four-ways or grids. It doesn't matter. All it needs is an Ethernet connection.

It claims it can harden grids, which are so fragile once they've got jobs and data running together, people are afraid to touch them lest the house of cards fall over, a situation that's not practical in, say, financial and pharmaceutical companies and so speaks against grids moving into the enterprise.

Back a year ago when Cassatt was Unlimited Scale, the clustering start-up formed by Cray and SGI refugees that Warburg Pincus asked Bill Coleman to look at, the idea was to build a version of Linux that could scale to manage thousands of Itanium and Xeon servers. (Think Cray-like MPP in software.)

Itanium has subsequently been abandoned in favor of the more popular Opteron and Intel EM64T. Cassatt now says that it will support Itanium if there's customer demand.

In Collage 2.0 every machine is fitted with an independent instance of Linux. Collage doesn't knit them together into a single-image system. There's no common memory. But Collage's single view trick makes them feel like one machine and during spikes in demand only what needs to be replicated for a job is replicated.

Apparently Collage 2.0 has been put into production at some unnamed manufacturing and financial houses. The manufacturer saw a 20 times improvement in price/performance over his old Sun rig with no changes to the scripting environment, McAllister said.

To get the product accepted, Cassatt has been working with solution partners like Kxsystems, Ascential, Informatica. Engineous Solutions and rival IBM. Apparently IBM has Collage in its Linux center.

Cassatt is now 100 people and has fleshed out its management team.

The great Rob Gingell left Sun a few weeks ago to sign on as CTO. He was responsible for so many things at Sun it's hard to know where to begin.

Suffice it to say for Cassatt's purpose that he was responsible for N1's underlying architecture, N1 of course being Sun's ambitious autonomic computing scheme.

Cassatt already picked off the Sun division in Colorado known as Rocky Mountain Technologies that was doing the systems management part of N1 and knows a lot about working with big logic. Cassatt relieved Sun of the whole team last year.

The start-up also recently hired Rich Green, Sun's Solaris and Java guy, and made him executive VP of product development. Then it brought in Sunir Kapoor, who started E-Stamp, the PC postage pioneer after stints at Microsoft, Novell and Oracle - he was responsible for Oracle's Collaboration Suite, one of Oracle's four main lines of business. He's now Cassatt's chief marketing officer.

Cassatt already had Mark Forman, who used to be the federal government's IT czar, a guy with a $60 billion annual budget, as executive VP, worldwide services, and of course Coleman, who turned SunOS into Solaris back in his salad days at Sun and then started BEA, which he grew to a billion dollars faster than any software company in history - including Microsoft and Oracle.

Guess he's gonna see if he still got the old magic.