24 Oct
2005
24 Oct
'05
10 a.m.
Huw Davies wrote:
Certainly the only 11/782 I know of (Melbourne
University) was split
to get much better performance. Basically all I/O (and hence all of
VMS) had to occur in the "master" 11/780 so all that could be run in
the slave was CPU intensive jobs. As soon as your CPU bound process
wanted to do something other than compute it had to be scheduled in
the "master".
The 782 certainly was two 780s joined at the hip. I thought that the
SBI was joined up too but that might be the Purdue (?) VAX-11/784.
When proper SMP was introduced in VAX/VMS V5.0 (the previous support
for the VAX-11/782 being dubbed ASMP), the 782 support was dropped
and the choices were either don't upgrade or split into two 780s.
Effectively you had one big global lock on VMS. The
story goes that
all the necessary VMS changes to support the 11/782 occurred over a
weekend. Another interesting story concerns the mythical 11/787 (ie
two 11/785s rather than 11/780s). There were rumors of such systems
occurring in a mythical site somewhere near the centre of Australia
There would have been no reason that a "787" would not have worked,
but given that bad rep that the 782 had, why would anyone have bothered?
Interestingly I don't think I've ever come across any technical docs
that describe the 782. It does rate a passing mention in the VIDSM.
The various "how we wrote support for the 782" articles that I've read
fail to mention that it was done over a weekend :-)
Somewhere or other I have Kathleen Morse's "how we might do
[A]SMP in VMS" memo. I think the memo itself was long enough
to take a weekend to write!
Antonio
--
Antonio carlini
arcarlini at iee.org