Re: Address bits again, Article on new mainframe use

On Wed, 4 Sep 2024 14:33:52 -0000 (UTC), John Levine wrote:

I think they thought it was paging, but of course 8K pages were way too
large.

Might have been to reduce the amount of hardware involved, and hence the
cost.

So they overreacted and the Vax pages were 512 bytes which were too
small.

The main issue, I think, was the proportion of physical RAM taken up by
physical page tables. VMS didn’t have to fault pages in one by one, of
course; it had the concept of a “page fault cluster size”, which was
settable in various places. For example, for executable files, it was a
field in the .EXE header; if this was zero (the usual value), it was
interpreted as 16. So paging in from the executable would usually be done
(as far as possible) 8 kibibytes at a time.

I remember when we got Macsyma (symbolic maths package, written in Lisp)
installed on our ’780, it was having way too much impact on system
performance. So our sysadmin patched the .EXE file, setting the page fault
cluster size to 1. And that improved things quite significantly.

Remember, that was going from an effective paging unit of 8kiB, down to
512 bytes.