On Feb 25, 2011, at 11:26 AM, William H. Magill wrote:

> As I recall, Virginia Tech's super computer (circa 2003) was 1000 G5s. (not 
> much info about it on-line anymore)

True . . .but didn't that predate the XServe?
> The biggest problem for Apple trying to run on Apple hardware is -- The lack 
> of a blade.
> 
> While the Mac Mini is tiny, it is still much larger than any blade 
> configuration. 
> There is a LOT of redundant "stuff" in a stack of minis equivalent to a 
> chassis full of blades -- speakers, disk drives, wifi, etc.
> Not to mention that the Mini doesn't support FibreChannel.
> 
> Which begs the question ... does any CONTEMPORARY Mac  support Fibre Channel 
> .... or have the slots to take FC cards?
> 

I'm not sure that the overall rack space for minis would be significantly more 
compact than a blade chassis. Granted; there are lots of different blade 
chassis . . .but if you figure a 6U or 8U rack unit that contains 16 blades (or 
even 32 for half height blades) . . .compared to the number of mini's that 
would fit in the same space I think you could fit more mini's since they're 
small front to back and could be stacked both side to side and front to rear. 
There is some power supply duplication as well . . .and I haven't compared the 
Xeon's in a blade to the Core 2 Duo in a mini . . .but it isn't intuitively 
clear that the blade chassis would provide significantly more processing power.

The lack of fiber channel in the mini would probably be an issue. I would think 
you could  get a FC card for a Mac Pro . . .but they're really not very 
efficiently rackable.

I wouldn't expect Apple to use minis . . although XGrid might be a possibility 
the lack of FC in the mini and the rack inefficiency of Pros makes them 
unlikely. Strictly from a server standpoint . . .commodity server hardware 
running Red Hat or a modified Darwin kernel seems like the best solution . . 
.although given that they might want to eat their own dog food the virtualized 
Lion installations running on VMWare might be a good idea as well. They could 
even XGrid a bunch of relatively small VMs to spread the performance across 
multiple physical hosts and let all of the FC/SAN stuff be handled by the 
commodity hardware.



-----------------------------------------------
There are only three kinds of stress; your basic nuclear stress, cooking 
stress, and A$$hole stress. The key to their relationship is Jello.

neil



_______________________________________________
MacOSX-admin mailing list
[email protected]
http://www.omnigroup.com/mailman/listinfo/macosx-admin

Reply via email to