RAM Disks OK

[mike at brl-vgr]

From:      Mike Muuss <mike at brl-vgr>

If all systems were virtual memory systems, and you could buy as much
memory for them as you wanted (no hardware limits), then your comments
are correct.

However, let me cite some examples of hardware where VAST performance
improvements can be obtained by using RAM DISK, and where no alternative
is availible:

1)  PDP-11/70 with max memory.  4 Mbytes is all you can attach.
    Worse, any one user can only use 128 Kbytes (split I/D).  Period.
    Everything else *HAS* to go onto "disk".  So why not make disks faster?

2)  VAX-11 with max memory.  Until 6 months ago, when the new memory
    controllers came out, a 750 could only have 2 Mbytes, and 780 only
    8 Mbytes.  It's virtual, at least, so things work better, but
    things like "C" compilers like to make large sequential files, and
    hand them from program to program;  it's not clear that putting these
    kinds of things in virtual memory is much of an improvement -- you'd
    just wind up page faulting more.

    However, on a Purdue-Dual-780 with 20 Mbytes of memory, we STILL plan
    on having a BULK MOS unit around for /tmp.

There exist computers which have no real limit to the amount of real memory
you can attach, such as our Denelcor HEP SuperComputer.  It has
8 Mbytes of "slow" 400ns memory, and 20 Mbytes of "fast" 50ns memory.
With the potential for adding more memory than anybody could possibly
afford.  For this kind of machine, clearly nobody would buy a RAM disk;
the money would be better spent for more "slow" memory (which holds
the UNIX buffer cache, etc).
					Best,
					 -Mike Muuss
					  User of 5 BULK MOS "RAM Disks".