Multilane_Connector

Dal Allan endlcom at acm.org
Sun Aug 18 22:29:46 PDT 2002


* From the T10 Reflector (t10 at t10.org), posted by:
* "Dal Allan" <endlcom at acm.org>
*
Howdy all,

Bob Thornton asked me to forward this to T10, so enjoy your reading.

Dal

------------------------------------

Respond to: bthornton at fcai.fujitsu.com

  Subject:  SFF SSWG re: Keying for 4-lane Copper I/O Connector

During the SFF SSWG meeting at T11, there was discussion concerning the
requirement (or lack thereof) for various keying of the 4-lane that
would differentiate the interface between the following standards:
 
  10G Fiber Channel - FC-PI-2
  Serial Attached SCSI (SAS)
  Serial Attached ATA2
  10G Ethernet
 
The concern has been voiced by some that keying was preferred, if not
required, to prevent the possibility of "smoking" a machine should the
I/O connector be plugged into the incorrect port. When discussed at the
SSWG meeting, and at T11.2 Copper, there was head scratching as to what
was in these standards that would allow for the "smoking" gun, so to
speak.  As these systems are all based on some form of XAUI, it was
thought that this was not a relevant concern. Please note that we are
not referencing InfiniBand, as the selected copper interface provides
sufficient keying to insure that the interfaces are not intermatable
 
Alvin Cox of Seagate expressed the following:
If every interface required AC coupling, there would not be a problem
unless voltage levels go too high. The thing we ran across during our
discussion within the SAS group is that not every interface requires AC
coupling on both transmitter and receiver. This is where we may
encounter a smoke situation. There is also the concerns regarding SATA
external cables. The electrical requirements have not been defined yet
for this interface and I cannot predict what the outcome will be (nor
can I share the information at this time). A table of all the interface
characteristics would be a great starting point. It is definitely a
necessary one.

I have been chartered (I did not get out of the meeting quick enough) to
chair an meeting where we can define these issues that are relevant
between these standards bodies to resolve this issue. I would like to
propose a possible meeting (schedule not defined - looking at September)
for any and all interested parties to meet with the goal to develop a
chart defining some of the potential differences (or maybe find that
there are no real differences) of these standards concerning cable
interconnect issues.

I feel that this issue can be (hopefully) quickly resolved to the
benefit off all interested parties in a quick and timely fashion (you
can tell I have never chaired one of these meeting).

I am traveling out of the country next week, so will be unable to
respond to any questions until the week of August 26th.

Regards

Bob

  - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

     The following response came from Bill Ham 
     after Bob posted to the T11.2 reflector.

Hi physical folks,

As noted in the T11.2 meetings the risk of smoking something is not
mitigated by ac coupling except in the unusual case where a high d.c.
level exists.  The main risks of smoking something come from transient
events that are just as likely between interfaces for the same transport
as they are between interfaces of different transports.  The T11.2 group
took the position that it does NOT want keying as a standard requirement
because it provides:

   * a de facto n-furcation* of the volumes for any one transport with
     resulting loss of leverage
   * very significant complexity in the configuration, ordering and
     servicing dimensions where multiple transports are supported
   * false belief that keying will prevent connector engagement to a
     determined individual who intends to make that stubborn connector
     mate no matter what
   * false belief that keying significantly mitigates the risk of
     damaging something (it acutally increases the risk that both parts
     of the connector, including the very expensive to replace bulkhead
     connector, will be damaged due to attempted forced matings)

Not having keying in the standard still leaves open the option for
anyone who wishes to have a closed system to have whatever keying they
desire.

There were NO discenting votes in the T11.2 group on the point.  So if
parts of the industry decide that they want their very own key, the
Fibre Channel key will be the null key (i.e. none) - assuming of course
that folks in T11.2 do not change their mind.

*n-furcation is where the total market is divided by n; the most common
of which is bi-furcation where n = 2.

Cheers, Bill

*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at t10.org




More information about the T10 mailing list