[rescue] J90 on epay

[Dave McGuire]

On Saturday, March 1, 2003, at 08:48 PM, Joshua D. Boyd wrote:
>>    In mine you mean?  My J90 has three IOSs...two in one VME64 chassis
>> and one in a second.  subsystems.  Each IOS has an IOBB interface 
>> going
>> into the mainframe.  See below for more details on this.
>
> I think you might have shown me the innards of Doug's, not yours.

   Ahh ok.  They're almost identical, config-wise.

>>    As far as I'm aware, the IOS SS5s set up DMA transactions between 
>> the
>> peripheral controllers and the IOBB board.  What happens then is
>> unclear...I'm not sure if the IOBB board buffers the data then DMAs it
>> into the J90's memory during a second transaction, or if it's more of 
>> a
>> "passthrough" interface.
>
> OK.  Perhaps it is capable of operating both ways.  Specially designed
> periphs do a passthrough, off the shelf periphs require two
> transactions.  Or maybe it does passthrough in a manner that allows the
> use of off the shelf boards.
>
> I've heard reports of off the shelf cards being used.  In particular, I
> heard one report of someone hooking a serial keyboard and mouse up, and
> adding a frame buffer to use the machine interactively.  Well, this
> might have been a J90 specifically.  I just remeber it being some YMP
> compatible machine.

   The only non-off-the-shelf VME64 board found in J90s is the IOBB 
interface.  They mainly use Interphase boards.

   The only direct-attach framebuffers I've ever heard of being 
connected to a Cray were HIPPI-attached ones.  I'm still looking for 
one of those.  I missed one on eBay about two years ago, haven't seen 
one since.

> How much of the machine have you managed to figure out?  Enough of the
> machine code to be able to make a start of coding ones own OS?

   All but the specifics of talking over the IOBB interface.

   There's really not much motivation for putting a different OS on one 
of those machines, though.  UNICOS is *nice*...and the compilers just 
can't be beat.

> I had no idea that HIPPI would be installed as something other than
> VME64.  Is this for latency reasons, or is VME64 not as fast as I would
> have expected?  For some reason, I expected that VME64 would be faster
> than PCI/X.

   I don't know the spec offhand, but VME64 is *fast*.  I assume it's a 
latency issue but I'm really not sure.  Possibly a sustained I/O issue.

   Actually, now that I think about it...HIPPI is usually treated as a 
network interface with a VERY LARGE MTU.  In fact, there's nothing 
about HIPPI that requires it to have an MTU at all.  I wonder if the 
HIPPI-attached-to-the-CPU deal has something to do with swamping the 
IOBB or the VME64 bus and starving the disk subsystem for DMA cycles.  
That would seem to make sense.

> 68020s seem to be used a lot as FEPs.  They are cool little chips.

   Yes they certainly are. :-)  And being very CISC, GCC actually 
generates decent code for them.

> I wish I had more time hack on those older 680x0 machines.

   Then get a job doing embedded systems development.  68K processors 
are *all over* the embedded world.  I'm working on a simple DragonBall 
(MC68VZ328, 68EC000 core) design right now.

         -Dave

--
Dave McGuire             "I've grown hair again, just
St. Petersburg, FL           for the occasion."       -Doc Shipley