11/70 Loads

[utzoo!decvax!ucbvax!unix-wizards]

>From obrien at RAND-UNIX Tue Sep 15 18:32:30 1981
	Here goes:

	We ran an 11/70 under what might be called "Version 6" for a large
number of years here.  We had made many, many performance improvements before
we were able to run 30-35 users with our split I/D editor.  Most of those
improvements showed up in Version 7, but for some reason V7 runs much, much
slower than V6 and no one knows why.  (Any takers?)

	You should be aware that the Berkeley 2.8BSD system is actually a
Version 7 distribution, as is the so-called Cory distribution.  They run
about as well as our old V6 did in terms of number of users (but with more
memory on the CPU, admittedly).

	We never noticed that the NCP put a significant load on the system
in terms of CPU time, but of course the kernel and the daemon are fairly
sizeable.  Our 70 now has a couple of meg of memory on it, though, so that's
not as much of a consideration - most of the NCP kernel was text space, not
data space, so we didn't lose too many I/O buffers.  I don't believe we
ever took the step of putting the buffers outside kernel space, but as I
was not involved with 2.8BSD, I could be wrong.

	We think the VAX will support about the same number of users
(30-35) but don't really have enough lines to test that out.  We soon will.
All of our lines, on the 70 and the VAX, run (ran) at 9600 baud.  Of course
you won't get anything like good performance unless you run DH-11's or
equivalent (Able DMAX or SUPERMAX, etc.), since at 9600 baud your CPU will
just sort of go away if you use DZ's or DL's, which are interrupt-per-char
on output.

	I know of no one migrating UNIX to a DEC-20, since it's a word-
addressed machine and difficult to work with, byte pointers notwithstanding.
Most people enamoured of mainframes have just ported UNIX to IBM architecture.