Using MONITOR from a 9600 baud terminal I've attempted to measure some
of the loads placed on the CPU by the printers and terminals.  The
command used to gather the statistics was:

	MONITOR/INTERVAL=3/VIEW_TIME=3  MODES/ALL/PERCENT

which I would run for a few minutes after setting up the appropriate
conditions and then record the average percent of time spent in the
various CPU states.  I believe this gives a valid measure of the CPU
load from the various states.


				% CPU time in mode (average)

				Int	Kern	Exec	User	Idle
				----	----	----	----	----

Idle (except for measuring	0.5	0.1	0.0	0.5	98.9
	terminal)

LPA0 running on long job	2.8	3.0	0.6	1.3	92.3

LPB0 running on long job	2.4	1.9	0.4	1.0	94.3

both LPA0 and LPB0 running	4.6	6.0	1.1	1.5	86.8

1 terminal at 9600 baud		8.3	4.8	3.2	1.0	82.7

2 terminals at 9600 baud	17.2	9.2	6.0	1.4	66.2

5 terminals at 9600 baud	39.0	23.6	14.5	3.4	19.5

7 terminals at 9600 baud	44.9	30.8	17.7	3.9	2.7

1 terminal at 4800 baud		5.6	2.7	1.9	0.8	8.9

2 terminals at 4800 baud	10.1	5.1	3.3	1.2	80.3

5 terminals at 4800 baud	24.1	13.2	8.5	2.0	52.2

The terminal tests were made using VT100's (and in some cases ADM-3A's)
TYPE'ing a long file over and over.  The VT100's were all set for
JUMP scrolling so that their baud rate was not reduced by the XON/XOFF
done for smooth scrolling.

DEC claims that an 11/780 running VMS V3 saturates at 25K characters/sec
with DZ11's.  My figures show that at about 6K characters/sec the CPU
is fully busy (essentially no idle time), but I expect DEC's number
represents the absolute upper limit where the machine begins to spend
nearly 100% of its time on the interrupt stack service the DZ11's.
Additionally, I did not change the terminal line characteristics from
their normal setting (except for speed) whereas DEC might be talking
about terminals set for /PASSALL.  Also, my test resulted in a range
of output I/O operation lengths (DEC appears to have used constant
block sizes).

As a reference, DEC claims the DMF32 raises the saturation point of the
780 to 250K chars/sec.  However, I also find a note that says DZ11's
are faster (about 60% of time) on small QIO's than DMF32's???  Need
more information from DEC on terminal performance using DZ11 and the
DMF32 before I can be really sure that DMF32 will greatly improve
things.

About the line printers.  With both going the CPU is spending 10.6%
of its time in Kernel mode and on the Interrupt Stack.  My notes
from DECUS indicate that the DMF32 would (if fitted to both printers)
only reduce this to 75% of its value or roughly 7.8% of the CPU
would be required to support the printers yet.  Obviously to justify
buying DMF32's we have to do it via terminal performance and we
need more information there.