🚀 go-pugleaf

>From twisto.eng.hou.compaq.com!wotan.compaq.com!cs.utexas.edu!uunet!haven.umd.edu!wam.umd.edu!rsrodger Wed Dec 16 15:10:33 CST 1992

>In article <1992Dec15.205639.25591@super.org> rminnich@super.org (Ronald G Minnich) writes:
>>In article <1992Dec15.194637.10009@eng.umd.edu> sysmgr@king.eng.umd.edu writes:
>>>You guys ever thought of licensing TurboChannel from DEC and building a
>>>Pentium box/server around it? 
>>
>>TurboChannel is a dog. 
>>
>>Actually, seems to me they could look at SCI. Except they want low cost.
>>But if the CMOS chips for SCI pan out, maybe they will meet the cost 
>>range.
>
>	How about QuikRing (Apple)?  
> 
>	<No flames, please--all I've seen is the hype and none of the
>	commentary--is there anything there besides hype and computer
>	mags babbling happily?>
>

QuickRing is Apple's highspeed I/O solution and it is a derivative of
SCI.  Our rational behind Anet is very much in-line with SCI guys',
except we wanted a solution that will be
A. for I/O (we don't want to carry the coherency bag)
B. useable for portables, desktops and servers
   (that means very few devices to a LOT of devices, consistent
   physical layer of add-ins and scaleable performance [see next bullet])
C. standard reasonably high performance and scaleable without changing
   the physical interconnects (an I/O device should work anywhere within
   the I/O boundary regardless of the system type)
D. cheap for system OEMs, IHVs and eventually for the users
   (that means high-volume, low pin-count, minimize burdens, leverage
   off today's low-cost technologies and utilizing future technologies
   should minimize any "damages")

I think SCI is neat for CPU/memory interconnects and I have a lot of
respect on SCI guys to come up with IEEE P1596. 

In the PC arena...

We have seen "fixes" such as the Data Translation bus on PC addin cards
to achieve higher bandwidth by going around or adding another
"high-speed" path to an existing path.  I think QuickRing is a quick fix.
Don't take me wrong... I think DT bus and QuickRing made sense for a lot
of people because they want to leverage off existing installed base.
Fixing an old inefficient solution is not really a crime.  But, what's
next when today's fixes run out of steam in the future?

I think it hurts to witness we all adopt something new, e.g., PCI or VL,
because we need more (bandwidth) but soon find out it-ain't-enough and
"fix" it by adding another path.  A case in point, PCI or VL bus peak
BW, as they stand today, will not be able to handle heavy video BW.
What's your guess on how you can achieve more BW without changing the
physical interface [see previous paragraph for hints]?

We looked at existing solutions including wide buses, point-to-point
and fiber links.  None of the existing solutions met our criteria.  So,
we started with a clean sheet of paper.

We want to start off with something sufficient for today's needs at
minimum cost and don't want to add significant cost in the future when
we need more.  We also want to minimize multiple levels of protocol
changes like today's solutions - having buses for CPU, local (PCI),
standard low-speed I/O (ISA), propriatary highspeed I/O (whatever),
standard medium speed I/O (SCSI), etc.  On the other hand, we also
want to offer for those who insist on having a "migration path".  We
want to offer and have options to optimize the I/O subsystem for
different criteria, such as, cost, performance, form factor,
expandability, compatibility, etc.

Based on what I know on QuickRing(QR)...
QR controller needs at least 78 signals (6 each for Up/Down and 33
each for Receive/Transmit ports).  The polyimide flex cable media
has lower dielectric constant than FR4 PCB, allowing the clock
rate at 200 MHz with sub-nanosecond edge rates.

My questions are
  - How much does a node cost including the controller, an impedance-
    controlled connector and the polyimide flex cable?
  - How many QR devices can I attach?
  - What is the QR board form factor?
  - How do I add an option card with a QR attachment?  Disconnect the
    ring, plug in the add-in card and connect the ring back with an
    additional ring segment?  How do I achieve hot-plug?  How do I
    take care of the card spacings and cable lengths so that they won't
    look too ugly outside the box?  Or do I need to take off that box
    cover?
  - How can I swap a QR device between an Apple and a Company-X?

We think we found a way to achieve A, B, C and D (see 1st paragraph)...

Is it dangerous to go with other solutions when there is a solution
that offers similar or higher performance and a good expansion story
at lower cost?

Hope this helps,
Kevin

[What's the next best thing when you can't give your customer
 the highest performance and unlimited expansion system for free?]