see my responses, prefaced PK:, interleaved
with your list of concerns.
1300 East Lookout Drive
Richardson, TX 75082
Geoff Thompson <gthompso@xxxxxxxxxx>
08/26/2008 01:40 PM
Please respond to
Geoff Thompson <gthompso@xxxxxxxxxx>
Re: [802.3BA] 802.3ba XR ad hoc next
At 06:35 PM 8/25/2008 , Swanson, Steven E wrote:
I think the 1/12 failures would
only occur if all links were 150m. If you believe the Flatman survey, 90%
of links are less than 100m so 90% of the links will work all of the time
with a standard that is designed to support 100m. Even if the other 10%
are all 150m, the failure rate is less than 1% overall. I think the number
less than 100m is smaller than 90% but even if it was 80%, the failure
rate would be less than 2% overall. And this also assumes that EVERYTHING
is worst case. If one always designs for worst case, we incur unnecessary
costs and place our customers in a difficult position.
That is not the way I read this.
My read is:
- Approximately 100% (99 44/100+%) of the links that are less than
100 meters will work all of the time.
They are the mass of the market and they are NOT the topic
Given the relatively small market for links that are between 100m and 150m
long, we have the following questions:
- How big is this market compared to the sub-100 meter market
PK: It's small relative to sub-100m market,
which is by far the largest of the markets we have. But even so,
it's a lot bigger than the market for 100G 40km SM optics, and may be bigger
than the 10km SM markets too. Yet you are not raising the same issues
around those PHY/PMDs. That is a strange inequity that I ask you
to explain. How big is the SM market relative to the sub-100m market?
- What are the odds that just using 100 meter hardware will just
work given the various margins and the miracle of statistics?
PK: a near guarantee, given the statistical
analysis methodology that is available to us. Yet I know very well
that if the standard does not explicitly say the optics are capable of
the longer distance normatively, it will be cause for system vendors to
not support it.
- How much of a premium could vendors demand over 100 meter hardware
to satisfy this market (at the same performance assurance level as the
sub-100 meter market)?
PK: no premium would be warranted using the
statistical approach, as exactly the same optics are used throughout.
- How much of a premium would the whole market tolerate to lump
this market with the sub-100 meter market?
PK: A moot question with the statistical approach.
- What could be done to mitigate this problem without changing the
hardware (e.g. informative annexes)
PK: The statistical approach is all the "mitigation"
one needs because the hardware is identical.
- What percentage of this market can be handled by various other
PK: A solution based on exactly the same PMD
as envisioned for sub-100m channels that can also support up to 250m leaves
virtually no part of the data center market not addressed and fulfills
the customers needs, rendering other solutions nearly pointless. But
of course vendors are always welcome to offer other alternatives, if they
feel it is worth the investment. With
the statistical approach no further investment is needed.
And when all of these questions are answered, we will be left with the
If we don't draw the line at 100 meters, where do we draw it?
PK: Data from several contributions shows that
drawing the line at 250m strands virtually no population.
The question DOES NOT GO AWAY when we move the line from 100 meters to
It just changes:
- The numbers from 100/150 to 150/something else
- The populations on each side of the hash
- The costs of the device
PK: The cost is nil with the statistical approach.