Phoronix Test Suite Brings Benchmarking To Mac OS X
Reply With QuoteInteresting. And more interesting as I installed OSX86 recently on my notebook (dell vostro 1400 with nvidia 8400m gs) and a desktop (wolfdale e8400 @ 3.6ghz on ep43-ds3l and nvidia 8500gt) on some spare hard disk space.
Although I don't quite like the huge size of the test suites. Pretty happy with small tests like Xbench which are just a few MB. (Btw, Xbench rates the above mentioned desktop close to 200, afaik meaning twice fast as best g5 workstation.)
I am guessing that the graphics subsystem tests in OSX will be pretty mediocre compared to Linux, especially those with ATI graphics. Just my guess though.
Aside, speaking of OSX, I guess the Linux distros can learn a number of nice things. First of them is the lighting fast boot time. I apparently see that it boots in barely 15 seconds or so. And its no magic as we can clearly see whats going on behind the scenes by booting in verbose mode (-v). The actual kernel is too small and most things are modular. Further, the modules (aka extensions) are sort of precached (Extensions.mkext) - not much of hardware detection goes on during boot. And it caches them automagically every time any changes are made to the extensions folder (and guess for hardware changes also). The other thing to notice is the small number of services and the way they are started parallely. The system feels very very responsive in general (also look at the output of 'ps -A' for the relatively small number of services running). Minimal and clean system I should say. Remember its also a server OS. I am sure Linux users don't want a 5-second boot (would be wonderful though), but 15-second boot is really possible and already achieved.
There is also another nice thing I saw about their kernel. Looks like there are not too many API changes. A hardware driver module (extension) made 2 or 3 years ago (for 10.4 I guess) still plugs in nicely and magically in 10.5.5. Thats something really really nice. Its kind of annoying in Linux that we can't expect third-party drivers to work with a minor change like kernel 2.6.26 to 2.6.27, that too for something as trivial as the number of arguments changing for a basic function! While its reasonable to ask for example that why fglrx developers cannot fix their drivers for xorg-server 1.5 which shipped with fedora9 several months ago, its also reasonable to question the number of changes made in kernel and xorg. Instead of chasing a moving target, devels can spend their valuable time implementing features and fixing *real* bugs - remember everything takes time. True that we see radical improvements in kernel and xorg, but things can hopefully be done in a little more non-destructive manner. Worst more, we don't want things like gallium vs dri2, gem vs ttm and ext4 vs reiser4 etc to be happening. Although these issues have been discussed time and again, hopefully somebody is listening.
Long one!
Originally Posted by hdas
