John Fowler, the recently appointed boss of Sun's network systems group, is in charge of cooking up an affordable systems strategy that enhances Sun's profits and its image. He's off to a good start by focusing on Opteron and Solaris.
The 64-bit Opteron CPU is the perfect successor to 32-bit Sparc. In Fowler's words, AMD's flagship processor gave Sun's engineers the power to "uncripple" Solaris for non-Sparc systems. A recent update to Solaris 9 improves Opteron's performance in 32-bit mode. The 32-bit Solaris 10 in testing is stable enough for limited production use, Fowler said. When Solaris 10 goes 64-bit on Opteron, it will run a mix of 32-bit and 64-bit Solaris and Linux apps, without performance penalties, on Sun servers and workstations with one to eight processors.
Fowler and Sun get my blessing for choosing AMD's technology over Intel's, but just choosing Opteron isn't enough to give Sun a leg up. In press releases, HP and IBM speak as though they came up with the idea of putting Opteron in servers, and AMD will not help Sun deflect their rhetoric.
As I see it, Sun must take two steps to use Opteron to competitive advantage. The first step would be to make Opteron Sun's sole platform alternative to Sparc, a level of commitment that Dell, Hewlett-Packard and IBM can't afford to make. And to make that one-sided monogamy work, Sun would have to pour a lot of software engineering resources into driving Linux deep into Solaris' shadow.
If Sun's engineers spend as much time porting Solaris down from 64-bit Sparc as they do porting it up from 32-bit x86, Solaris 10 will shame Linux in speed, agility, and manageability on Opteron in 64-bit mode. Sun's got the code base and the engineering chops to pull it off, provided they don't take the shortcut of treating Opteron like a Xeon with more memory. The word from John Fowler is that Solaris 10, Sun's C and Fortran compilers, and Sun-branded Java 1.5 VM will make Opteron scream. He's seen what Sun can't show yet, and he's jazzed in a well-dressed geek kind of way.
Sun can engineer its butt off and still not win back its throne. Unlike Solaris on Sparc, Solaris on Opteron has an installed base of zero. That forces vendors to choose between Novell, Red Hat, Windows and Solaris as targets for their AMD64 and x86-64 ports. I put Sun fourth in this contest because ISVs remember that Sun let its original Solaris x86 starve to death.
If boasts and promises were the same as execution, Sun would win every game. Maybe Solaris on Opteron will be Sun's straight and true road south of Sparc. Or maybe Sun will just make a lot of noise about that pairing in July and go back to waffling among Linux and Solaris and Xeon and 32-bit Sparc so that no one -- customers or software partners -- can make sense of it. It will take an engineer to solve this one, and not just any engineer. Mr. Fowler, I hope your cape is back from the cleaners.
Yager is the technical director of the Infoworld Test Centre.