solid-state paging...

[utzoo!decvax!ucbvax!unix-wizards]

>From decvax!utzoo!henry at Berkeley Sat Dec 26 22:59:59 1981
Putting "10 pounds of processes" into a "2 pound sack (memory)" by keeping
(you hope!) inactive processes out on disk is PRECISELY using the disk as
an extension of main memory.  I.e., it uses disk to pretend that there
are 10 pounds of memory there.  The differences between paging and swapping
are details of how big a lump is moved around and when.

The replies to my first letter seem to have missed my basic point:  if
you want to attach a new lump of memory to your machine, it is clearly
better (other things being equal) to put it on your bus directly rather
than on the other side of a DMA interface that makes it look like a disk.
Directly-connected memory is more flexible and there are fewer hassles
in using it.  Directly-connected memory CAN be used as an exact equivalent
of a solid-state swapper, if you insist;  performance will be better
because the swap-in and swap-out operations are no-ops involving no setup
and no i/o cycles.

In reply to vax135!jfr, if reasonably-sized memories are not on the
market, what is inside the solid-state paging boxes?  If physical or
architectural limits prevent attaching available large memories to your
bus, THAT is exactly what I cited as the one real use for a solid-
state paging box.  As for a 100Mbyte memory, since existing VAXen have
a tendency (I'm told) to run out of CPU before they run out of memory,
I greatly doubt that you can buy a machine today that can USE a 1GB
memory intensively enough to justify it (unless it is running something
really strange like a 700-MB process with real-time response constraints).
And if the only difference it makes is pushing the hit ratio from 99.99%
to 99.999%, why bother?  You'll never notice the difference.