[Info-vax] RealWorldTech on Poulson

glen herrmannsfeldt gah at ugcs.caltech.edu
Tue Jul 5 10:21:19 EDT 2011


Johnny Billquist <bqt at softjar.se> wrote:
> On 2011-07-04 17.54, Michael Kraemer wrote:

(snip)
>> That was the Mips R4000.
>> And given the RAM constraints of that time,
>> 64bit addressing was almost useless.

> By the way, 64 bit addressing was not that useless. RAM size isn't the 
> only thing here. A VAX has 32-bit address space at a time when machines 
> shipped with 512K, and something like 4 Megs was considered huge. The 
> large virtual address space still made a big change on how you could 
> write your software.

Actual 64 bit addressing is useless, and will be for some
years now.  More than 32 bit is, but we are still many years
from being able to use up a 64 bit virtual address space.
(Terabyte disk drives are currently readily available, so 40 bits
of virtual address space.)

IBM S/370 and VAX have a two level virtual address space.
IBM has segment and page tables, VAX has pagable page tables.
That is necessary, such that enough of the page tables can be
kept in memory.  IBM z/Architecture has a five level virtual
addressing system to cover all 64 address bits, but with convenient
ways to "cheat" and avoid some levels until the whole address space
is needed.  (Otherwise, it would require five memory references
for every address translation not in the TLB.)

Starting with Pentium Pro, Intel went to a 36 bit physical address
space, along with the 45 bit virtual address space available since
the 80386.  If they didn't put a MMU with 32 bit addressing in
between, IA32 would have lasted much longer.

-- glen



More information about the Info-vax mailing list