Having said nothing for months about its intentions, a deafening silence that had Sun users biting their nails down to the quick about where the technology was really going - then demonstrating their angst in their order rate or their leap into the waiting arms of competitors - Oracle the other day finally opened its mind a tad.

For starters, it staged a webcast Tuesday and said that it means to put its resources behind Sun's Sparc and x86 systems, something it said before however unconvincingly. But now it's got what it called an "aggressive" albeit stripped-down - and for that matter pretty much unreadable - five-year roadmap to wave around that plots out Sparc server development through 2015.

The plan is to tightly integrate Oracle applications with Sun systems software and hardware.

To tickle application performance Oracle plans to double Sparc performance every two years and bring a 64-socket, 128-core server to market in five years time that supports 64TB of memory.

That 2015 system is supposed to be capable of running a whopping 16,384 threads, up from 256 today.

On the way there, there are supposed to be high-end Fujitsu-made Sparc64-based M series chips and more power-efficient volume T series chips. Sun's 16-core Rock chip was canceled months ago.

Fujitsu staying in the picture for the next five years remains a question. John Fowler, the former Sun hardware executive now running Oracle's systems business, didn't even confirm whether the T3 Rainbow Falls Sparc chip will make it out his year like it's supposed to although that seems to be his basis of most of his comparisons.

Unlike Fujitsu there is no AMD issue. There will be no Oracle AMD boxes. Oracle's x86 efforts will go into Intel blade and rack servers, an intention that proves Sun ain't Sun anymore. In view of its x86 volumes, the new Sun is sacrificing its old politics for a better margin and Intel's currently superior per-core performance. Oracle's licenses are based on how many cores you've got not on how many sockets.

Of course Oracle's hewing to Intel could change when AMD's new Bulldozer microarchitecture arrives.

It's also got plans for a next-generation Solaris 11 operating system next year. It'll be the first major release of the Unix OS in six or seven years and is supposed to scale over time to thousands of threads and tens of terabytes of memory.

Virtually the whole operating system will be overhauled with immediate attention paid to maintenance, security and file systems followed by annual updates. The end result should be a system that runs OLTP apps 40 times faster - figure 120 million data transactions a minute - in part a function of the new Sparc chips.

Before that, Oracle will jig the widgetry's high availability, memory scalability and virtualization then its I/O scalability and systems management and finally its threading.

A Solaris 11 early adopter program is supposed to kick off this year.

Fowler made no mention of OpenSolaris now in the process of forking out of sheer exasperation with Oracle's studied neglect.

All the Sun widgetry is supposed to run Oracle's software better than IBM, HP or Dell machines.

Fowler put in a good word for the Red Hat-derived Oracle Enterprise Linux (OEL) on x86 and recommitted to Oracle Virtual Manager for running Windows and Linux. Oracle's management console is supposed to bridge Oracle and Sun's heterogeneous x86/Sparc virtualization widgetry and cover up their differences. Sun's OpsCenter will be Oracle's provisioning tool for physical servers and integrate with VM Server and Oracle Enterprise Manager.

Oracle recently signed up HP and Dell to resell Solaris x86, OEL and the Xen-based Oracle VM along with Oracle support. IBM means to stop selling Solaris x86 later this month.