RE: [EFM] RE: [EFM-P2MP] Point-to-Point plus Shared Media
my comments in-line, after the ">>>"
--Dave Horne
-----Original Message-----
From: carlosal@xxxxxxxxxxxxxxxxxx [mailto:carlosal@xxxxxxxxxxxxxxxxxx]
Sent: Friday, November 23, 2001 10:13 AM
To: Roy Bynum
Cc: Horne, David M; 'Angeloni Cesare, IT';
'stds-802-3-efm-p2mp@ieee.org'; 'stds-802-3-efm@ieee.org'
Subject: Re: [EFM] RE: [EFM-P2MP] Point-to-Point plus Shared Media
David and Roy,
First of all, a disclaimer. I've been working with the group for some time,
and as such, I've listened to many different proposals for bandwidth
allocation on the Ethernet PON. I would like to make it clear that I'm not
associated with any vendor, nor do I have any commitment to a particular
technology, algorithm, or trade secret related to this particular topic.
I've been working with the requirements subgroup, and I'm supporting their
presentation. So please take this as a independent opinion on the technical
merits of the proposal; I'm not going to enter into any debate if it's time
or not to present new ideas, protocols or algorithms to the group.
>>>OK, but I'll offer my thoughts on that matter :o). When you consider that
we have no consensus market studies or deployment cost models to back up the
viability of any of the approaches presented, and that a consensus decision
has not yet been voted on, I would hope that we are all open to methods
which reduce cost and complexity, and have higher performance. This
translates into larger appeal. Hopefully a standardization effort considers
marketability and performance first, and artificial market windows last.
First mile fiber deployments will last for decades, and are manual-labor
limited in terms of deployment rate, so near-term market windows are
artificial.
There are some differences between what I called 'stateless' bandwidth
allocation protocols and 'stateful' protocols. David's proposal belongs
with the second family, as every ONU has to keep track of state information
to synchronize its transmissions.
One potential problem with 'stateful' protocols is that they're prone to
stability problems. If any member of the network lose a single control
message, you're set for trouble. For instance, it can happens if
connectivity is briefly disrupted at some point; it also can happen if a
ONU goes down unnoticed, effectively 'starving' everyone that is waiting
for it's transmission to end.
>>>That is simple to solve with a default timeout. I only addressed the
basics of "normal" operation in my earlier post; it wasn't intended to be
all-encompassing.
Also, stateful protocols pose some security
concerns; a single forged control message has lasting effects over the
network.
>>>Headend can command everyone or anyone to stop at any time, and restart
them in any order, in any combinations. Forged messages are a problem on any
shared medium network, not just the proposal I made. It is less of a concern
with the small sizes of PON segments, than compared to 500-2000 on a cable
nwk. It is also less concern on a optical link since a greater level of
expertise is needed to couple anything into the fiber, compared to a copper
wire.
Stateless protocols minimize these problems by making sure that the status
information is short lived, and is frequently refreshed. This helps to make
sure that every ONU will always have a recent copy of the control
information. Any instability is short-lived; at the next control
communication cycle, the information is refreshed. Stateless protocols also
tend to adapt faster to changing conditions; at every control cycle, the
allocation can be fine tuned. So I would suggest that any further
development of this idea to consider making the control message short-lived
- a single round would be fine.
>>>The method I proposed is capable of varying this to whatever the network
operator is comfortable with, and I tried to describe that in my earlier
message. I don't think this interval would be mandated by the spec; just the
capability is mandated. It is up to the individual network operator, and
they can vary it with load if they want. That said, if I were a network
operator I would tend toward the frequency you describe, and leave it
static.
Another potential development would be to use the ranging information to
sort the transmission sequence. By this method, stations closer together
would transmit in sequence, thus increasing the performance of the network.
Simulations can tell us if the performance gain is significant.
>>>Agreed, but the gains seem pretty incremental for the level of complexity
that would be added. Also, an accurate simulation model might be tough to
arrive at. A contrived case could easily be created, but how representative
would it be is the question. I'm open to discuss any detailed suggestions
you had in mind.
Carlos Ribeiro
CTBC Telecom