Sun 3/110 memory problems, VME adaptor cards

[Charles Hedrick]

I agree that 4MB of memory is no longer enough for any system that is
going to use window software.  We're going to try to upgrade all of our
systems this summer.  However I find that NeWS 1.1 works OK under 4.0 if I
recompile it under 4.0.  Using the shared libraries saves just enough
memory to make a difference.  Also, someone from Sun comments that you
should make sure that you are runnng a kernel that is properly configured
for your system.  Appendix E (?) of the Read Me First for 4.0 gives
details on doing this.  I've gone even further, and removed support for X
(it was version 10, and never really worked) from NeWS, as well as
recompiling some modules with GCC, which generally produces tighter code.
(This is all on the Sun 3.  I'm not sure whether I trust Sun 4 GCC yet.)
With all of this, NeWS works acceptably.  There are still times when there
is noticable paging, e.g. when I do a compilation in one window, the
compilation tends to cause the window stuff to page out, so when I want to
type something to a window there's a pause of a couple of seconds while it
pages in.  But I can now go between psterm windows, and between psterm and
emacs, without the long pauses due to swapping that I used to see.  I
don't see much difference between NeWS and X in this area, by the way,
though for some kinds of tasks X is faster.  (The main one is reading
news.  It puts up new messages much faster.  Based on the speed of Emacs
with NeWS support builtin, I think NeWS itself is capable of putting
things on the screen quickly, so I believe some more performance work in
psterm is called for.)  Like you, I shudder at the thought of running
merged X/NeWS on a 4M machine.

About memory for 3/50's, there are now 3 vendors: Clearpoint, Sunflower,
and Helios (?).  Helios (if I've got that name right) requires some
soldering on the board, not of components, but to mount the daughter board
solidly.  Our field service people obviously don't want us to do that.  So
I'm ordering one from each of the other two vendors to evaluate them.  I
understand that at the most recent Sun users' group meeting, this issue
came up a number of times, and Sun said that they didn't intend to do
anything about the 3/50 memory.  Furthermore they said that they wouldn't
support your systems if you got memory from someone else.  I hope my staff
member was exaggerating, because from his account, Sun was incredibly
arrogant: "why should we help someone else make money?"  (To the
moderator: Did McNealy really say that?  How did he get out of the room
alive?)  The answer is obvious: if your customers can't get more than 2
years of useful life out of a machine, when they replace it it won't be
with one of your products.  We think that at the local level, Sun field
service will be more understanding.  (There are, after all, plenty of
other companies interested in servicing our machines.)  Indeed I know a
number of people within Sun who are very excited about these memory
expansions, since it gets them out of a very difficult bind: they are now
shipping software which their hardware won't in any reasonable sense run.
It may not be true of their president, but most people within Sun do
realize that it's better for them if their users can use their systems.

Both Clearpoint and Sunflower require you to remove several chips and put
funny connectors in their sockets, however this is (in principle)
reversible, assuming that you use the right tools (which they supply),
take static precautions, and are careful in doing chip extraction.  The
static precautions are essential, since you'll be dealing with sensitive
chips.