[plug] Serial ATA

Craig Ringer craig at postnewspapers.com.au
Sun May 18 00:35:52 WST 2003


> I think that PCI express thing is part of this program at Intel to get
> rid of the legacy (what we are using now that apparently hasn't changed
> since the 286) PC architecture. 

Believe so, yes. I gather that one of the interesting PCI Express 
features should be the possibility of having a PCI card on a /cable/. 
Think card in drive bay or disk with integrated and private 
ATA-controller. Or, as an extension of that, HDD drivers *arrggh*.

OK, lets leave off the paranoia. PCI Express looks significantly 
different, where PCI-X is just "fast PCI". Very fast. The server at work 
can take PCI-X but I can't imagine what I could put in it that could 
soak a 133MHz/64bit bus.

> It includes other wacky stuff like having a sector of your HD setup with
> BIOS type info instead of having a bios.I imagine this will have
> advantages and dis-advantages but I dunno what they expect us to do for
> thin-clients. I tend to like that I can test a PC even if it doesn't
> have a (working) HD.

Yeah, because it worked /so/ well for Compaq.

EFI looks interesting, but I don't trust Intel to do a good job. I'm 
sure it'll be perfectly engineered - but I'm not sure it'll be 
engineered to ensure that the user has total control, or have 
undocumented bits that make it hard for other chipset manufacturers to 
make completely compatable products.

I seem to remember there is a spec/proposed standard on EFI, and I 
gather there is a "sample implementation" available ... but we'll have 
to wait and see in the end.

The other thing to think about: BIOS virus! If you're keeping a FAT32 FS 
on the disk containing BIOS data, "drivers", etc - its a real risk, 
though no doubt there will be efforts made to hide that data from the 
OS. Also  just merrily swapping bootable disks from PC to PC could get 
... "interesting".

As for thin clients - it isn't at all hard to see the use of, say, a 
32mb ATA flash module for storing the EFI, etc. Hell, it might prove 
possible to implement things like an X server straight in EFI, and cut 
out the OS entirely. Intel's documentation on it made it look pretty 
full featured (networking, host FS access, device drivers, etc) and if 
they're ditching "legacy" mode it certainly won't be single-task 16bit 
code.

I for one will be /delighted/ to have a PC that ditches all the legacy 
crap - no more virtual ISA for stuff in the south bridge (serial ports, 
etc) ... no more crappy keyboard controller ... no more workarounds for 
memory space design assumptions ... no more ugly boot code that has to 
guess what the BIOS is thinking ... the possibility of an easy, built-in 
multi-boot system ... the list goes on. Maybe some of this won't happen 
but I gather most will (from my reading of Intel's material on the 
matter) and if it does - I'll be very happy. If I want to play Master of 
Orion 2, I'll do it in an entirely virtual machine with my ridiculous 
amount of CPU power, or I'll find a P100 to run it on. Backward 
compatability has a cost too, and there can be a time to drop it.

I just hope they do it right, and don't try to pull anything slimy.

Craig



More information about the plug mailing list