What I found really interesting is how Apple is working on what I call "infrastructure" issues in this OS X release over a relentless "glitzy features uber alles" agenda. The seemingly mundane issues they're going after tell me that they're going to go after the business market through servers.

There's only one OS out there now if you really want to deliver maximum performance from SMP multi-core or multi-CPU platforms: Sun's Solaris. That's been true for years now, since Solaris 2.5 or so. (in 1998 or thereabouts). Windows, Linux, FreeBSD -- they've all done OK with a couple CPU's, but they all fall down (quickly) once you get to four CPU's or more. Multi-processing looks a lot easier than it is. When you start trying to deliver an OS that will deliver a nearly 1:1 scalable increase in usable CPU and memory bandwidth as you add CPU's... suddenly things get a little more complicated than they first look. For example, in Solaris 2.1, you could add a second CPU, but after you went to 4 CPU's, you found that processing got slower the more CPU's you added. This was a result of large chunks of the kernel being under one spinlock - any time you did IO, you ended up locking down one code path on the first CPU to do the IO. Doing really scalable IO processing on multi-CPU platforms is very fiddly stuff.

From what I gather from Apple's announcements, they're going to try to deliver threading & SMP processing on par with Solaris on multi-CPU/multi-core platforms, which will give them an edge on back-room servers (eg, racks full of Xserve blades). Adding reduction in bloat and footprint... almost makes me want to work for them. I can't tell you how many times we begged for a release where we could "stop the feature flood" and just clean up what we had already shipped.

The idea of using GPU's for user-land processing... that's a neat hack too. Lots of compute bandwidth is available in GPU's, especially if you're a business and not a gamer.