8 Oct
2003
8 Oct
'03
9:36 a.m.
Patrick Finnegan wrote:
.
I've just recently got a VAXstation 3520
Something I'd love to see up close someday...
[ More about installing additional CPU boards ]
That would be interesting - looking forward to comments from
Antonio and others about whether this is really possible...
On a related note, I there's a MicroNote about running multiple
MicroVAX II CPUs (KA630, M7606) in a single backplane. Up to 4
in fact (MicroNote #26, on www.ibiblio.org). Has anyone on the
list ever worked with, built, or seen such a beast? Did any
interesting OS research projects of the time use this feature
(I'm thinking V or Mach)?
--S.
8 Oct
8 Oct
10:43 a.m.
Steven M Jones wrote:
On a related note, I there's a MicroNote about
running multiple
MicroVAX II CPUs (KA630, M7606) in a single backplane. Up to 4
in fact (MicroNote #26, on www.ibiblio.org). Has anyone on the
list ever worked with, built, or seen such a beast? Did any
interesting OS research projects of the time use this feature
(I'm thinking V or Mach)?
Check the note again. AFAIRC, this (ka630, ka650, ...) had just a
"mailbox" register to communicate, and only one cpu had access to the
qbus. The other ones had the qbus interface shut down. So, if you didn't
have you special backplane with 4 qbuses, it was pretty limited ...
cheers
10 Oct
10 Oct
12:19 a.m.
On Wednesday 08 Oct 2003 8:43 pm, emanuel stiebler wrote:
Steven M Jones wrote:
I've been planning a monster like this, 4 small backplanes mounted in a
hacked-around wide cabinet, 4 processors running as a VAXcluster over
Ethernet...
Been done before?
alex/melt
On a related note, I there's a MicroNote
about running multiple
MicroVAX II CPUs (KA630, M7606) in a single backplane. Up to 4
in fact (MicroNote #26, on www.ibiblio.org). Has anyone on the
list ever worked with, built, or seen such a beast? Did any
interesting OS research projects of the time use this feature
(I'm thinking V or Mach)?
Check the note again. AFAIRC, this (ka630, ka650, ...) had just a
"mailbox" register to communicate, and only one cpu had access to the
qbus. The other ones had the qbus interface shut down. So, if you didn't
have you special backplane with 4 qbuses, it was pretty limited ... 
9:03 a.m.
I've been planning a monster like this, 4 small
backplanes
mounted in a
hacked-around wide cabinet, 4 processors running as a VAXcluster over
Ethernet...
If you mean 4-way multiprocessing, then 2-way was almost
certainly done at some time (probably in DECwest). I'm not
sure 4-way is practical, at least not with 4 backplanes. If
you want them all in one backplane you probably have to remove
(or otherwise inhibit) the bus terminating resistors on 3 of
the processors.
As others have pointed out, it doesn't really get you anywhere since
there is no real shared memory between the processors (they could
each have their own via PMI or they could all share some Qbus
memory - neither is an appealing option). You would also have the
minor problem that no OS would be able to do much with the
slave processors. (In fact, I think all the Oses I know of
probably wouldn't even notice them, so they would probably
boot as though they were on a MicroVAX II (or III if you used
KA65x processors).
If OTOH you meant 4 separate BA23 boxes each with its
own processor all in a VAXcluster, then that's just
independent MicroVAXes ... I had racks like that in
the lab at DEC. But that's less exciting than pushing
the boundaries I guess. You don't need a hacked cab,
just an ordinary 19" rack will do ... I kept back-to-back
VAX 4000s or MicroVAX IIIs in the bottom
to save space and make *absolutely* sure the rack
could not be tipped :-)
Antonio
--
---------------
Antonio Carlini arcarlini(a)iee.org
9:28 a.m.
Alex White wrote:
Why bother doing you own backplanes, if you use ethernet anyway ?
If you're thinking about you own box (doing your own backplane), try to
use four ka655 with 64 mbyte memory each and 4 mbyte shared memory on
the qbus.
But again, probably no software :(
Check the note
again. AFAIRC, this (ka630, ka650, ...) had just a
"mailbox" register to communicate, and only one cpu had access to the
qbus. The other ones had the qbus interface shut down. So, if you didn't
have you special backplane with 4 qbuses, it was pretty limited ...
I've been planning a monster like this, 4 small backplanes mounted in a
hacked-around wide cabinet, 4 processors running as a VAXcluster over
Ethernet... 
11:35 a.m.
"emanuel stiebler" <emu(a)ecubics.com> wrote:
If you're thinking about you own box (doing your
own backplane), try to
use four ka655 with 64 mbyte memory each and 4 mbyte shared memory on
the qbus.
AFAIK, the KA650 and KA655 do not have the support for multiple
processors on Qbus that was present in the KA630. The KA630 (MicroVAX II)
had two features for this:
1) The bus interface could be configured to be a bus requester rather
than the system bus arbiter. This has to be done on all CPUs other
than the first. (Note that the VAX 11/780 was also capable of this,
as discovered and exploited by George Goble).
2) Each processor has a "doorbell" register to allow other processors
to interrupt it.
12:47 p.m.
AFAIK, the KA650 and KA655 do not have the support for
multiple processors on Qbus that was present in the KA630.
The KA630 (MicroVAX II) had two features for this:
1) The bus interface could be configured to be a bus requester rather
than the system bus arbiter. This has to be done on all CPUs
other
than the first. (Note that the VAX 11/780 was
also capable of
this,
as discovered and exploited by George Goble).
It may be, as you say, that the KA65x cannot do this. The KA655 TM seems
to be silent here - or at least it is in the sections I quickly read
just
now. The bus interface is handled by the CQBIC, but I don't have that
particular manual handy right now. If the CQBIC can do this, then I
expect
that most of the CVAX MicroVAX 3000 Series machines can do it (and if it
cannot, then presumably none of them can).
2) Each processor has a "doorbell" register
to allow other processors
to interrupt it.
This I think it does have - at least it has a register called IPCR. I
don't
have the KA630 User Guide (*sniff* - someone must have one they can
scan!)
so I cannot compare terminology directly, but I think this is what the
doorbell register is. Since I don't know of anything that uses it,
finding
example code to verify this may be somewhat hard!
--
---------------
Antonio Carlini arcarlini(a)iee.org
3:41 p.m.
On Fri, 10 Oct 2003, Eric Smith wrote:
1) The bus interface could be configured to be a bus
requester rather
than the system bus arbiter. This has to be done on all CPUs other
than the first. (Note that the VAX 11/780 was also capable of this,
as discovered and exploited by George Goble).
Don't forget that Mike Marsh also was involved in this, and did most of
the hardware work on the "Dual-VAX". As someone who has the pleasure of
working for him, it seems like he doesn't really get enough credit for
being the co-inventor of SMP Unix...
Well, that's my $0.02
Pat
--
Purdue University ITAP/RCS
Information Technology at Purdue
Research Computing and Storage
http://www.itap.purdue.edu/rcs/
8 Oct
8 Oct
12:14 p.m.
"Steven M Jones" <classiccmp(a)crash.com> wrote:
On a related note, I there's a MicroNote about
running multiple
MicroVAX II CPUs (KA630, M7606) in a single backplane. Up to 4
in fact (MicroNote #26, on www.ibiblio.org). Has anyone on the
list ever worked with, built, or seen such a beast? Did any
interesting OS research projects of the time use this feature
(I'm thinking V or Mach)?
The resulting system does not have shared memory (except for any
Qbus memory), so it isn't generally suitable for operating systems
that support SMP. That's one reason why DEC never officially
supported such configurations.
12 Oct
12 Oct
12:44 p.m.
That would be interesting - looking forward to
comments from
Antonio and others about whether this is really possible...
I *think* that, although only the 3520 and 3540 made it out
as officially available systems, that you could in fact
build up at least a 3560 and 3580 (i.e. up to four CPU
boards yielding a total of 8 CPUs). That's just based on
having seen something in the VMS Listings a long time
ago.
There was also a speed bump considered, and possibly
built, which got as far as being allocated a system number,
possibly along the lines of 38x0 (i.e. 3820, 3840 etc.) and
that one had a tantalising 38A0 as a possible name IIRC.
This speedup for the Firefox (codenamed Fastfox) never
made it out of the door. I expect that multicpu workstations
were just not in vogue at the time.
I presume that the reason that the 3560 and 3580 were never
marketted was down to performance scaling badly with
inceasing number of cpus, or possibly there may have
been a fatal problem (like the 3-CPU VAX 8300 variants
occasionally locking up and the 4-CPU variant being
pretty much guaranteed to lock during boot).
Antonio
--
---------------
Antonio Carlini arcarlini(a)iee.org
12:49 p.m.
New subject: VAX 8600 on ebay
http://cgi.ebay.co.uk/ws/eBayISAPI.dll?ViewItem&item=3051901405&cat…
=1479
It does say "too large for pallet" so I guess you have been warned. This
is
really a souped-up ECL VAX-11/780, and I know that several of you are
quite
smitten by that vintage of machine. (sadly for me it's on the wrong side
of the pond and has appeared just[1] before I become rich).
Plenty of docs can be found on the net already, but no printsets (that
I know of). Looks like this one might come with printsets (since it
has maintenance docs) so would whoever buys it please get them scanned
:-)
Antonio
[1] "just" here being measured on a geological time scale :-)
--
---------------
Antonio Carlini arcarlini(a)iee.org
7590
days inactive
7594
days old
10 comments
6 participants
participants (6)
-
arcarlini＠iee.org -
classiccmp＠crash.com -
emu＠ecubics.com -
eric＠brouhaha.com -
meltie＠meltie.com -
pat＠purdueriots.com