More ENABLE/34 Info...
utzoo!duke!decvax!ucbvax!Mike at BRL
utzoo!duke!decvax!ucbvax!Mike at BRL
Sun Jul 12 22:58:31 AEST 1981
From: Mike at BRL
To satisfy the requests for more detail, I have abstracted some of the
more interesting parts from a paper entitled "Modifications to UNIX to
Allow Four Mega Bytes of Main Memory on a 11/40 Class Processor" by
Clement T. Cole and Sterling J. Huxley of Tektronix Laboratories:
The ENABLE/34 PARs are indexed by bits 13-17 of the Phys Addr comming
out of the CPU. 5 bits => 32 PARS can be used; 8 for Kernel I, 8 for
Kernel D, 8 for User I, 8 for User D. The DEC PARs have to be set up
to map into different parts of the 18-bit address space. Roughly:
KD0-6
KI0-7
UI0-7
UD0-7
KD7 I/O page
is the 18-bit layout, allowing the 32 ENABLE PARs to be used to really
do the management. The DEC PARs are intialized once and forgotten.
If you hope to use Supervisor Mode, you loose!
Mapped Memory Access time is quoted as Memory access time
plus 150 ns. Cache hits access in 110 ns max.
Fortunately, DEC did not assign any other I/O devices resiging at
the I/O map's location for non-22-bit machines, so the ENABLE Unibus Map
appears identical to the DEC 11/70&44 one.
No speed quotes are given, so you probably also suffer the 90-100 ns delay
for NPR address mapping, same as the 11/44&70.
"Unfortunately, it was found that the ENABLE board can not
be running in 22-bit relocation mode with I/O mapping turned OFF.
These two bits should be independent of each other..." == Hardware glitch.
A software kludge can get around this.
Just to show something, Clem ran some benhmarks, which due to time
limitations, can not properly show off the benefits of the ENABLE.
The benchmark is a recompilation of their UNIX Kernel...
With all that extra memory, a 60 buffer block cache in UNIX would have
made worlds of difference...
CPU Real User Sys
================================
11/70 13:06 5:03 3:12
11/44 23:35 7:42 6:39
11/34 58:23 17:12 13:08
34+ENAB 42:19 14:42 10:29
These results I suspect as being rather pessimistic -- see the paper for
details.
Putting the Able 8K cache on an 11/34 makes a big difference.
This remains true for the ENABLE/34.
For those interested in prices:
ENABLE/34 -- $5,500
ENABLE CACHE-$3,750
and you should pay no more than $4,500 for a single board 0.25Mbyte 22-bit
memory board.
I know nothing more about the hardware -- call Able at (714)-979-7030 and
pester them. If need be, ask for Ken O'Mohundro ("The UNIBUS Wizard").
Personally, I think that this is the shot-in-the-arm that our aging 11/34s and
11/40s needed, for those of us who don't have the luxury of replacing all our
11's with VAXen.
Happy Hacking!
-Mike Muuss
More information about the Comp.unix.wizards
mailing list