Cringely’s Machinations

Saturday, 22 April 2006

Continuing on a theme from his previous two columns, Robert X. Cringely is still predicting that Apple will replace the Mach kernel in Mac OS X 10.5 because it’s slower than the kernels in Linux and FreeBSD. Mac OS X is slower than other leading Unix-style OS’s for server-based tasks, and I believe the Mach kernel is the main reason for this, but it’s not that much slower for graphical-user-interface-based tasks, and that’s really what matters most for Mac users.

It isn’t ludicrous to speculate that 10.5 (or 10.6, or some subsequent update even further out in the future) would feature a new kernel — Mach is definitely not one of the best parts of Mac OS X, and the kernel is so deeply under-the-hood that most users would never notice. But I don’t think it’s likely. What’s likely is that Apple will continue improving Mach’s performance incrementally, just how they’ve been doing from 10.0 through 10.4.

(Also, just forget about Cringely’s explanation about “integer calculations” being the cause of the performance difference. That’s not it at all, and the real reasons are densely technical. Trust me that it has nothing to do with “integer calculations”, which is a claim that doesn’t even really make sense. Check out this snippet of a thread from the Linux kernel development mailing list in which Linus Torvalds writes, “I claim that Mach people (and apparently FreeBSD) are incompetent idiots,” if you want a taste for the sort of issues that cause Mach to lag performance-wise. Update: I point to this not because I think Torvalds is the final word on kernel design, but simply to show that serious arguments regarding the merits of Mach compared to Linux are deeply technical and in many ways beyond the ken of laymen, yours truly most definitely included.)

Anyway, what caught my attention is Cringely’s further speculation that this new kernel will be for Intel-based Macs only:

Speeding-up performance is great, but normally a system vendor won’t want to do that for older hardware, which might encourage some users to keep their old box and just add a new OS. But in this case, Apple HAS NO installed base of Intel Macs to worry about having to compete with, so speeding up the OS becomes a no-brainer, especially if it simultaneously encourages PowerPC owners to upgrade so they can share in the fun.

For this reason alone, I’m guessing that the new OS X Kernel won’t be backward compatible to Power Macs. But this is just a guess.

That wouldn’t make any sense at all. If 10.5 has a new kernel, and the new kernel doesn’t run on Power Macs, then it would mean that Mac OS X 10.5 wouldn’t run on Power Macs. There is just no way that Apple is going to ship PowerPC and Intel versions of Mac OS X with two entirely different kernel architectures. And there’s just no way that they’re not going to support PowerPC Macs fully in 10.5 — there’s just too much upgrade revenue to be made. ¹

Plus, Apple has been increasing system performance on older hardware with each successive release of Mac OS X — 10.1 was way faster than 10.0; 10.2 was noticeably faster than 10.1, etc. If Apple can make 10.5 noticeably faster than 10.4 on PowerPC hardware, they won’t hesitate to do it. Because whatever hypothetical kernel improvements might make 10.5 faster on PowerPC hardware, they’d make just as big a difference on Intel-based Macs, and Intel-based Macs are inherently faster.

I.e., these first-generation Intel-based Macs are so fast that Apple doesn’t need to worry about suppressing performance on PowerPC Macs to spur people to upgrade to new hardware. The Intel-based Macs are selling themselves.

One more intriguing tidbit from Cringely’s column:

Now for the interesting part: I believe that Apple will offer Windows Vista as an option for those big customers who demand it, but I also believe that Apple will offer in OS X 10.5 the ability to run native Windows XP applications with no copy of XP installed on the machine at all. This will be accomplished not by using compatibility middleware like Wine, but rather by Apple implementing the Windows API directly in OS X 10.5.

[…]

I’m told Apple has long had this running in the Cupertino lab — Intel Macs running OS X while mixing Apple and XP applications. This is not a guess or a rumor, this something that has been demonstrated and observed by people who have since reported to me.

This strikes me as wildly ambitious — it’d be a fantastic achievement, technically. Cringely cites Apple’s and Microsoft’s 1997 five-year patent cross-licensing agreement for why this might be possible: Windows XP shipped in 2001, while the agreement was still in effect. But patent cross-licensing is a far cry from source-code cross-licensing.

Perhaps the hardest part about building a new implementation of an existing standard such as the Windows application APIs is that you not only have to achieve feature compatibility, you also have to achieve bug compatibility — a Windows app running on Mac OS X would need to behave exactly as it does on Windows. That also means implementing not just the documented APIs but also the undocumented ones, and the fact that they’re undocumented makes them rather hard to implement.

The ongoing Vista fiasco indicates that Windows is so large and so complicated that even Microsoft is having trouble producing a new version of it.

Running Windows itself in a compatibility layer — à la the way the Classic environment runs Mac OS 9 — would be a lot less work on Apple’s part, and would almost certainly be a lot more compatible with the zillions of Windows apps out there.

The difference with Cringely’s idea is that with his scheme, every Mac running 10.5 would be able to run Windows software right out of the box, not just those Macs on which a licensed copy of Windows is installed. But there’s no reason for every Mac, or even most of them, to be able to run Windows software.

Licensing fees for Windows aren’t necessarily skin off Apple’s back — Apple can either continue with how they’re doing it now with Boot Camp (i.e. “bring your own Windows”) or they can make pre-installed Windows a build-to-order option at the Apple Store, which would mean that people who wanted it would have to pay an additional $100 or so.