- by Paul Murphy, -
I've been puzzled by Sun's Linux strategy for a long time. On the software side things make sense: Linux is Unix, supporting Linux in the short term brings new value to Solaris in the long term. On the hardware side, however, I couldn't see any logical reason for Sun to sell x86 boxes until just last week when Intel announced their 64bit Xeon extensions. My idea about this is pure speculation, of course, but if I'm right Sun may be creating some interesting opportunities for Linux users everywhere to save a few bucks now while positioning their companies to lead the next great wave in technology change.
There are several layers to this, so the place to start may be with what's so troubling about having Sun sell x86 hardware. That's pretty simple: they can't make money at it. As HP's latest quarterly report shows they can't either, and it's fairly clear that IBM subsidises every x86 sale as a loss leader on services. Dell makes money, but probably more because its management made some very smart decisions during the Asian currency melt-down in 98 then because of their vaunted internet marketing model.
It's not just that margins are tight on this stuff, it's how things got this way that really counts. Commoditization cuts both ways, with standardisation allowing a shrinking group of suppliers to dictate costs to retailers like Dell and HP while raising the volume bar for others trying to enter the business. Consider, for example, just how close these companies are on pricing for a typical "industry standard" PC server - say something like a rack mountable, dual processor, Xeon with 8GB, a pair of internal 72GB US3 disks, Linux, and three years of "Gold" support.
The two most closely comparable machines in this category seem to be the Sun v65 and HP ML370 G3 rack version. Both of these support dual Xeons, up to 12GB of RAM in eight slots with a 533MHZ FSB, six internal 1" US3 disks, and six PCI-X cards. Preloaded with 8GB, 2 x 72GB US3 Disks, Red Hat Linux, and three years of support these come in, using the vendor's enterprise pricing as of Feb 22/04, at about $15,389 for Sun and $15,592 for HP.
Dell's 2650 is a close match differing from the other two really only in its limitation to five PCI cards of which only three may be PCI-X. Otherwise comparably configured, it costs $15,335, or $54 less than the Sun. Neither Gateway's 975 Rack server nor IBM's x335 are directly comparable, with Gateway's entry being larger and shipping with SuSe Linux instead of Red Hat's at $15,767 and IBM's allowing only two internal drives, 8GB, and two PCI-X slots for $16,567 when maxed out.
New entrants to an existing market traditionally have to offer either, or both, additional value and/or significant savings. In this case Sun preloads Linux and is 1.3% ($203) cheaper than its most directly comparable competitor. That's not enough to get dedicated HP customers to switch, and even its $1,178 advantage over the less capable IBM machine isn't going to draw many converts from among the IBM faithful. Bottom line: some existing customers will buy them, but Sun can't expect profitable volume growth from x86.
So what's the deal here? if HP and IBM can't make money selling the same or lesser boxes for more, what reason is there to believe that Sun can? Scott McNealy used to describe x86 retailing as adding bruises to bananas, but now he's doing it. Why? It's certainly not because a chorus of financial analysts thinks it's a great idea; these people have an enormous influence on the share price, but their actual technical sophistication and performance as forecasters would be unacceptable for the likes of Punxsutawney Phil and Buckeye Chuck. It's not because there's a lot of technical synergy either: Sun really does design and build most of its own SPARC machines, but it's getting those x86 units from the same people Dell, HP, and IBM buy them from. So the bottom line here is simple: Sun can't expect to make money selling these; it can't expect to use them as loss leaders to steal customers from Dell, HP, or IBM; and, it can't expect to build synergies from doing this into cost savings in other areas.
In other words, x86 is a loser for Sun just as it would be for Apple and for the same reasons. Both companies already have better products of their own and sales organizations far more capable of delivering those products to willing customers than of breaking into the dog eat dog world of nickel and dime PC retailing.
So why do it? I think the reason was Itanium and the battle for all those user sites, an estimated 700,000 of them, being forced off Tru64, VMS, and HP-UX. The Itanium started out with a pretty good design idea and some research challenges they had every reasonable expectation of beating. For SPARC a truly successful Itanium would have been bad news, so going from merely hoping that AMD would repeat its success against the Pentium Pro to actually helping them do it would not have seemed a big step back in 2001. Look closely at the Opteron today and you can see some serious Sun design influences starting with the much higher memory bandwidth and going all the way to a set of RISC like instructions that allow pages to be marked as execute only - thus preventing things like simple buffer overflow attacks.
If Sun wanted to help AMD build a market for Opteron, it could obtain significant synergies from selling a bit of Xeon stuff early on in the process, particularly if the first few hundred thousand go to an existing customer base where Sun has a cost of sales advantage. Doing this lets Sun frighten the PC retailers into pushing Intel on 64bit extensions while developing its own supply and delivery channels and rapidly reaching the quanitities needed to get lower contract pricing from assembly operators in places like Taiwan. That makes sense and is, I think, the top layer to the motivational onion here.
The second layer, however, is both more speculative and more fun. Treat the Opteron/Itanium issue as stage dressing, you know the gorgeous distraction on whom all eyes focus while the magician does the real work in the background and you get a different picture.
That real work isn't just to help AMD kill off Itanium, but to take advantage of the processor vacuum this creates to considerably extend Sun's market. How? Well, Intel's announcement of its 64bit Xeon extensions is missing two things: the bandwidth potential in the AMD design and the RISC like security features. Taken together this means that a multicore capable Solaris on Opteron will have significant performance and security advantages over any OS running on Intel's upgraded Xeons while maintaining compatibility with Linux on any x86 product, whether from AMD or Intel.
Making that advanced technology a bit cheaper than the Xeon stuff should help that along too. That's why, I'm guessing, Sun's 64bit V20z, with 8GB of DDR1-333 ECC Memory, two 72GB disks, 2 PCI-X slots, and dual opteron M248 processors running SuSe Linux costs $12,846 - $3,721 less than IBM's dual Xeon.
That combination of high performance with low price and industry standards gives Sun a powerful selling story to tell users whose systems, whether VMS, Tru64, MPE, or HP-UX, are being dead ended by HP. They can cut short term costs by moving rapidly to Sun's Opteron boxes while regaining their position at the leading edge of new technology adoption --and do it without the risk of being burnt again.
As a sales pitch it's not only cool, it's true. Of course, the other part's true too: once they're inside the Sun tent -a place most Alpha users would otherwise never go- they'll soon see that Solaris on SPARC actually provides more bang for the buck. And that, as they say, is the money bet.