Aucbvax.1984
fa.unix-wizards
utzoo!duke!decvax!ucbvax!DEDWARDS@USC-ISI
Mon Jun 29 17:24:37 1981
Heavy stuff on the IMP11-B!!
FYI - about the IMP11-B.....

The only good thing that I can say is that the IMP11B does work, but....

It is KMC-11 based which is really why it is so bad.  The KMC is
a real bear to program and has timing problems with the UNIBUS to boot!
When you start up cold in the morning, the IMP microcode has to
first be downline loaded.  That is no big deal - but the timeing
problems are.  On the 11/70 especially,, there is a 
for(i=0;i<1000;i++) line in the driver because the IMP11B says that it is
ready to send out the next packet, but REALLY it ISN'T!  You
end up wasting lots of time in the kernel because of this whenever
packets are xmitted to the net.

In addition, the interrupt struucture of the KMC is to be kind,
convoluted.  Instead of a xmit and rcv interrupt, there is
a transmit command and receive command interrupt structure.
This means that when a packet is sent or when a packet is received,
the receive command interrupt gets tickeled and then the driver
has to go through a skip chain in order to decode the type of interrupt
(i.e., transmit packet completed, packet rcved from net, imp
going down, ...).  The other interrupt is used when setting up a
command to be given to the IMP11B - i.e, tell the IMP11B to read
a packet from the net.  To do this, the IMP11B has to be told
that a command sequence is coming.  The IMP11B then has to respond
either by rasing a bit (you wait in a loop until the bit comes high
which is the way my driver works) or you take an interrupt hit.
Then, the actual command is given to the IMP11b!  WHEW!

Also,  when the IMP11B breaks, DEC typically does not
know anything about them!  Luckily, we have several spare sets of
both KMC-11s and the IMP11B line cards.  I have had cases
here the DEC diagnostics have said that the IMP11B was fine,
but UNIX would not come up,  where the diagnostics said all
was OK and UNIX would come up but the NCP would not talk to the net.
Usually our solution is to swap boards and see if the problem goes away.

At any rate, yes, the IMP11-B  does work.  My driver can be made
available, the firmware downline loader too.  But, on our new 11/44, we have
an ACC LH/DH-11!!!

If you want the driver just to look at, let me know and I can
mail it to you.  I also have a paper on the subject that I could
mail to you (its kind of long so US mail might be better) if you
would like.  Finally, call if you want anything cleared up
(301/796-6380 or 6345).

One more note, I do hear that DEC had an engineering hold on IMP11-B's
which may mean that they are serious about trying to at least fix
the bus timing problems.  The KMC itself is beyond help!

Howard Weiss
(weiss at TYCHO or dedwards at ISI)


-------