More on 11/70 Loads, and Swapping on an RK05

utzoo!decvax!ucbvax!unix-wizards utzoo!decvax!ucbvax!unix-wizards
Wed Sep 16 11:10:03 AEST 1981


>From mike.bmd70 at BRL Wed Sep 16 11:03:43 1981
Mike -
	Yes, we really swap on the RK05.  It turns out that having
an entire controller dedicated to the swapping helps more than the
reletively slow transfer rate of the RKs hurts.  EXEC arg lists go
out there, and pure text prototypes.  We don't do enough real swapping
(except for core shuffling) that it hardly makes a difference.
'Tis better than taking up seek time on the moving head disks,
which are usually saturated with I/O.

	The ROOT is accessed far more frequently on our system than
is swap space .  I have seen over 200 blocks/sec xfered from the root
to user processes, and still had the moving head disks busy.  For us,
it is a big win.

	Put some more memory on your 11/45 with an ENABLE/34, and
forget about swapping...

	We have put a lot of work into our scheduler.  It certainly seems
to be much fairer than the Bell Round-Robin approach!  Keeps wasted cycles
to a minimum, too.
				Cheers,
				 -Mike



More information about the Comp.unix.wizards mailing list