From: Spagna,
Fulvio
Sent: Tuesday, September 28, 2004 4:28 PM
To:
Altmann, Michael W
Subject:
Minutes 9,28,2004
Signaling ad hoc
(September 28, 2004)
- Meeting called to
order.
- Mike reviewed ad-hoc
work items. Focus on :
- Aggressors
treatment (e.g. random variable etc.)
- Channel test case
selection (tyco channels, synthetic channel etc.)
- Discussion opened on
treatment of channel selection:
- Ali
John has
brought variable data. But not all possible combinations are perhaps not
feasible/realistic.
- Mike
Are you
proposing that we standardize some form of driver impairments for driver,
package etc.
- Ali
The
problem is that the current models have a lot of reflections which previous
models did not have.
- Mike
This is a
positive direction if we want to consider an end to end simulation. We need
some proposal for what to do with respect ot this impairments. Did you have
something in mind Ali
- Ali
Yes I can
come up with some typical s-parameters for package and ESD as a starting
point.
- Richard
Why typical and not worst-case ?
- Ali
Because it may not work.
- Richard
Then
there is something wrong with the channel?
- Ali
No, if you use everything at worst case nothing will
work.
- Charles
Just to give a direction to this. We are trying to do a signaling direction,
no trying to ascertain over what range the system will work. We just
need to have a large enough samples so that we do not have unfair bias
toward a signaling method and not another.
- Joe
Should
the sample be restricted or should we also include some
outliers.
- Charles
It was actually your presentation that showed that channels
in-spec behaved worse. But I think it’s reasonable to say that we
should exclude the models that do not meet the
template.
- Mike
However
we had a discussion in a previous meeting in which we reached a different
conclusion (to evaluate the level of margin).
- Jeff
You are right that you should pick cases that are representative of
real life but that meet the template. We, as signaling group, should not
augment the population by adding other cases.
- Mike
The
original purpose was not to provide a simulation set that the signaling must
pass, but rather provide a bench to evaluate the relative performance. This
is way real life channels were requested and
provided.
- Steve
Vendor #2 (which required very little equalization) may provide additional
data under the same conditions Tyco did.
- Mike
A
That is fine, we had originally talked about a maximum of twenty models, we
are at eight so we are fine.
- Mike
L.
I agree with Jeff that the channel ad-hoc went through a lot of work to set
a template and we should restrict ourselves above that line. Joel has gone
through considerable effort to build test cards and I do not see his data as
part of this suite of test cases.
- Mike
A
So you would exclude channels that would violate the
template.
- Mike L.
Till that channel ad-hoc decide to move the template we should not move it
for them. We should increase the suite of test cases. Joel and Intel are
potential sources for additional models …
- Joe
… although none of the Intel ones fall within the
line.
- Mike
A
So we need a specific proposal.
- Ali
How do we treat proprietary models …
- Adam
… these are not to be used to define compliance but only to compare
signaling …
- Ali
…
- Adam
… so give us a proposal.
- Charles
I would
like a straw poll to accept John’s 7 models, Steve synthesized models,
Joel’s models
- Mike
We already did, except for the latter. At the time we explicitly accepted
the idea to have some models below the template
line.
- Joel
I have easily 12 more models
- John
I can also provide more models. And these are the test cases you requested.
Everyone of those channels was requested and now you want to reverse
yourself.
- Mike
This is an engineering effort and just because something was requested three
calls ago does not mean that they are needed.
- Ali
This is the first we saw of these models …
- Mike
A
… not true.
- Brian
We can certainly ask for a superset.
- Mike A
How do we get to a concrete proposal. (Joel hands the chair a list of
channels, #1,2,3,6,7,8,14,17,18). OK, we have nine channels. Joel, would it
be possible to graph them.
- Joel
I would need the list back …
- Some vivacious
discussion ensues on the subject of the backplane template and on whether we
should exclude channels that violate the templates.
- Mike
A
There is a request for a straw poll to accept, as additional cases, Joel’s models #1,2,3,6,7,8,14,17,18
- Jeff
Did one of those have 3 connectors?
- Joel
we would have to add channel 19 but I am uncomfortable in including this as
it violates the template.
- Jeff
I just do not want this to drop of the edge of the
table
- Mike
Well, if that can be fixed it can be added
later.
- Bill
The problem with this data is that the data does not fit the goal of
the group (as being connector less and not 40” and not representative of
realistic channels)
- Joe
So that means it’s an easier channel.
- Glenn
Myself, wasn’t to get some kind of model for simulation. However, there were
at least three presentations in which it has been hinted that the
template will change. So what we may decide may be in vain.
- Mike
Well, back to the purpose (signaling performance evaluation etc.
etc.)
- Straw poll
(inclusion of Joel’s models):
- Adam
The reason this task force was formed was because there was no way to
compare different results.
- Justin
What I object to is to weight the results …
- Discussion erupts
once again.
- Mike
A
There will be channel discussions. But for signaling we need to start
somewhere, and this is our charter. If we do not get pass the selection of
channel models we will not be able to accomplish our
goal.
- Straw poll (should
we only consider for future inclusion models that do not violate all
the templates as set by the channel ad hoc):
- Mike L I
would like to propose a straw poll to revise the previous
models.
- Adam Why
should we not let the channel data drive the signaling selection. Later we
can decide the weight to assign to the cases that do not meet the template.
If we do anything in the signaling ad hoc contingent on the channel ad hoc
we will never start and have no data to work with.
- Meeting is
adjourned.
|