Subject: RE: Chapter 46: preamble length
Pat:
I agree with what is stated below, and would add the point
that this really would be a different kind of delete because
of running disparity issues within the data.
Thomas Dineen
Brad,
Since we seem to be taking a poll - I agree with Brad. Any
device causing
preamble shrinkage in 10 Gig Ethernet would be
non-compliant. All the data
manipulating physical layers (meaning those that handle
encoding and
decoding) have state machines specifying how the encoding
takes place and
those state machines show that when data is received, the
data is encoded
and sent. We have text that explicitly allows deleting idles
and, in some
circumstances, ordered sets. We don't have any text allowing
deletion of any
kind of data and the layers below the RS have no knowledge
of preamble.
WIS, the PMAs, and the PMDs do not do deletion.
Furthermore, we have made sure that there are plenty of
idles available for
deletion. On the other hand, preamble is only 8 bytes and
deletions can only
be done in 4 byte chunks. If devices were relying on the
ability to delete
preamble, there is only one chance per frame so if a device
higher on the
stack deleted one, there are no chances per frame.
If you believe that the current draft text allows deletion
of preamble, then
it also allows deletion of an arbitrary 4 bytes of data
because there is no
text in any of the PCS clauses that treats preamble
differently than any
other data byte.
Also, note that if a device deleted preamble, it would be a
different kind
of deletion. Normally deletion happens to a column of idle
or an ordered
set. Neither of the preamble containing columns is deletable
because one
contains the /S/ and the other contains the SFD. An
implementation deleting
preamble would have to delete some bytes from one column and
some from the
next column combining the ends of the columns to make a new
column.
Regards,
Pat