Apple should start taking enterprise servers seriously
Not as cool as iPods, but profitable
A range, with headroom, now
An Apple server needs the same kinds of features and options that you see in Wintel boxes today: a range of processor sockets and speeds, networking, disk controllers, disk drives, integrated tape and other media for backup and recovery, and system management tools. There should be tower machines aimed at SMBs with one and two sockets, and maybe as many as four sockets, with enough disk capacity to provide a few terabytes of RAID 5-protected disk capacity. The rack form factor that Apple uses for the Xserves is not appropriate for office environments, which is why Hewlett-Packard, Dell, IBM, Fujitsu-Siemens, and others sell lots and lots of tower servers. (And this is a lesson that has been lost on Sun Microsystems, hence it has not done as well in x64 servers as it had hoped.) Ditto for rack machines. There should be low-cost, single-socket, barebones machines, mid-sized ones like the current Xserve, and larger machines with four sockets. That means supporting the "Dunnington" four-core and six-core processors from Intel at the least. And maybe, just maybe, Apple needs to go as far as to support a line of machines built on Advanced Micro Devices' "Shanghai" Opteron processors.
I realize, of course, that two-socket servers are the bulk of x64 sales. But no enterprise customer wants to buy into a platform that does not have SMP scalability. They want headroom in terms of CPU, memory, and I/O capacity - even if they never actually use it.
The iron is the easy part. No question about it. Enterprise customers are not driven by coolness (unless they are roasting inside their data centers), but rather by their application choices. And this is probably why Apple has been less than adequate as a server supplier. But this is not an impossible task. For a modern server platform, there are a few dozen databases and bits of middleware that absolutely need to be certified to work on the platform. Vendors throw a party when a given Unix or Linux release breaks through a 2,500-application barrier, and it is not unusual for a release to top out at around 3,000 applications over the course of its life. The task is not insurmountable. If the key server platform providers can do it, along with the operating system suppliers Microsoft, Red Hat, and Novell, then Apple can do it, too. Let's face it: If Itanium can get application support, Mac OS X should be a relative snap.
And because VMware's ESX Server hypervisor has already been tweaked to virtualize the existing Leopard Server (Mac OS X 10.5), virtualization is not an issue. (The product is called VMware Fusion, but it is really ESX Server.)
Just to be clever, Apple could do something it already has the most expertise in the world doing: emulating applications from one platform on its current x64 iron. The "Rosetta" emulation environment, which allows Mac OS applications compiled for PowerPC iron to run on Intel-based Macs, could be used to move Unix and Linux applications over to Mac OS X. Of course, that would mean striking up a new licensing deal with IBM, which ate Transitive, the owner of the QuickTransit emulator that underpins Rosetta, last fall. And that might be a bit tricky, with IBM's Power processor expert, Mark Papermaster, in court trying to get his job as a tech guru for Steve Jobs over at Apple. IBM doesn't want Papermaster to move over to Apple, and it sure doesn't want Apple to jump into the server racket, either.
The other option, of course, is for an intrepid server vendor not afraid of litigation to hack Mac OS X onto their existing x64 server lineup. It has been done by wiseguys on desktop machines, but given that enterprise customers are running mission-critical applications and they like real tech support, this idea is not really tenable. But it is what Apple deserves, just the same. ®
Sponsored: Benefits from the lessons learned in HPC