Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index


Hi Gerry,

Sorry to bring such a discussion so late to the table. First of all, a

- I do not want to disturb the current discussions.
- I do not want to make us go back to the beginning.
- I do not want to waste energy in a discussion that is not going to lead
us anywhere.

Said that, let me explain the context of my inopportune suggestion.

As we start to analyze the MPCP protocol with more detail, it becomes clear
that it will be very difficult to achieve high efficiency on the upstream.
There are so many variables to account for: laser turn on/off times, guard
times, report message overhead, framing overhead, and so on. Small grant
sizes are important to minimize latency and make the system more
responsive, but OTOH, they make the system even less efficient. Part of
this is caused by the decision not to fragment frames, which was a very
wise decision, given the complexity of fragmentation.

Of course, the system can be made more efficient: tight timing, better
components,etc. This may add a lot of complexity to the system design. So I
was wondering if we could not take a different approach - instead of higher
efficiency, throw more bandwidth at the problem. It's a somewhat dumb
strategy, but it worked for several technologies, including Ethernet, for a
long time.

In my mind, the tradeoff is just like this (just an example, the figures
are arbitrary):

Which one is more cost-effective: a 90% efficient 1 Gbps system, or a 50%
efficient 2.5 Gbps system?

Of course, this is not to say that we need to start from 2.5 Gbps. We can
start from 1 Gbps, and then go to define 2.5 Gbps, or even 10 Gbps, as the
next step. This would allow us to select a less efficient but inexpensive
MPCP implementation. It's also one of the places where good wavelength
allocation planning can come up to save the day, to allow for easier

Sincerely, I'm not concerned with efficiency by itself. There are other
more important criteria: reliability, scalability, predictability,
responsiveness, etc. All (other) things being equal, what really matters is
the bottom line, and that's the cost per bit. If that's 1 Gbps or 2.5 Gbps,
it's really not important; we have no emotional attachment to powers of ten
anyway :-)

Best regards,

Carlos Ribeiro
CTBC Telecom