[rescue] 128 bits...

[Scott Newell]

>I think that may make sense up to a point, but 64 bit address space is
>a bit huge... I am fascinated by the idea of a CPU that addresses
>directly into secondary storage, data files exists as address
>locations, not inodes or sectors/tracks/platters on a SCSI ID...

I think it's been done--IBM's DASD, perhaps?  I'm sure someone will poke in
with a datapoint...

But why stop there?  Why not put the entire network into the address space:
all the memory, all the drives.  Dereferencing that pointer might be
spinning up a platter on the other side of the planet!


Adding 32 bits to the memory space: if memory prices fall by 50% a year,
that gets you there in 32 years at equal cost.  Assume $100 for 4GB now,
and you can have $100k full populated in ~10 bits less, or ~22 years.  Drop
memory prices to 25% each year, you fill it in ~11 years.  Granted, that's
a pretty long time span to extrapolate over.


newell