Draft Minutes, Parallel SCSI WG - March 7, 2000

Gene_Milligan at notes.seagate.com Gene_Milligan at notes.seagate.com
Tue Mar 14 08:42:27 PST 2000


* From the T10 Reflector (t10 at t10.org), posted by:
* Gene_Milligan at notes.seagate.com
*

<<The group discussed the degree to which data can be transmitted at rates
below
the negotiated rate.  The problem is that the current and previous versions
of
SCSI use the negotiated data transmission rate as a maximum -- transmission
at
slower rates are perfectly legal.  However, transmission at rates other
than
negotiated rate cause problems for the timing budget compensation functions
(skew compensation, adaptive equalization, etc.) that rely on training
operations because the training results are only good for the transfer rate
at
which the training was performed.  Everybody had been assuming that
transfers
would occur at the negotiated rate, which would also be the rate at which
the
training was performed.  The potential for slower transfer rates raised
questions about such assumptions.  One agreement was that, if slower
transfer
rates are used, the training must be done at the slower rate and redone
anytime the rate is changed.>>

     Well not quite:

<<However, transmission at rates other than
negotiated rate cause problems for the timing budget compensation functions
(skew compensation, adaptive equalization, etc.) that rely on training
operations because the training results are only good for the transfer rate
at
which the training was performed.>>

     I pointed out that slower rates should not cause problems for skew
compensation but that the slower rate does not, in a straight forward
implementation, increase the setup time, only the hold time.

<<Everybody had been assuming that transfers would occur at the negotiated
rate, which would also be the rate at which the training was performed. >>

     Not everybody but I gathered there were some that did and that was why
I pointed out that was not the case.

<<One agreement was that, if slower transfer rates are used, the training
must be done at the slower rate and redone anytime the rate is changed.>>

     I did not hear any such agreement. Certainly the skew compensation
training does not have to be done differently for slower rates. I assume as
a matter of course that a slower transmitting device, if they exist, would
transmit at a rate determined by their chosen clock frequency with setup
time nominally the same as for the maximum nominal and hold time larger
than the maximum nominal (maximum nominal so called before tolerances). So
then the question is whether or not anyone has to do frequency domain
training. The model assumes fixed clock rates in multiples per negotiation
but not otherwise variable except for tolerances for drift.


Gene








*
* 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