Thanks for the additional analysis.
It is not surprising that the removal of all links <10m from the
C-A subsystem did little to change the statistics of the overall portion
of links >100m because the numbers of remaining C-A links still swamps
that of the A-D and D-C links by 9:1.
That analysis adjustment misses a major
point, which is that one must treat the C-A subsystem separately from the
A-D and D-C subsystems. They serve different purposes, just as horizontal
cabling in office buildings serves a different purpose than backbone cabling.
Consider this. In a commercial building analysis, one would
not mix the horizontal and backbone links together in an attempt to determine
what the customer needs. It would skew the result to the short length
end of the distribution because horizontal links, which are generally shorter
than backbone links, outnumber backbone links by more than 10 to 1. If
we followed this misguided approach we would conclude that commercial buildings
have little need for anything but copper media that serves up to 100m.
We know this is not true. The same thing applies to the data
The difficulty for some in seeing this
is perhaps that the mental model of a building is easier to parse, given
all the years of structured cabling models and standards, than that of
the data center. None the less, the structured cabling model of the
commercial building has been applied to the data center, albeit with different
names in some circles. The horizontal is the C-A subsystem, and the
building backbone is the combination of the A-D and D-A subsystems.
We can argue over whether the right
percentage of links >100m in the data center backbone subsystems is
10% or 20%. But the fact is that any of these percentages they are
significant, especially when the solution offered to support them is so
much more costly than that for <100m. Perhaps one way to make
an adjustment for the magnifying effect of the differential in cost is
to multiply the percentage of longer links by the cost differential factors.
After all, for the cost of the SM solution one could install more
than two 40G XR links and more than seven 100G XR links. This serves
to show that like the short links, the longer links too must be well served
with a low cost solution so that they do not represent a market acceptance
Regarding the misreading of the "intent
to deploy" rate as a "penetration rate" in your survey report,
I think that the intent to deploy rate may be a reasonable surrogate for
the ratios of deployed links in the various subsystems. After all,
if a customer is deploying 40G/100G in the C-A subsystem, they would certainly
need it in their A-D and D-C subsystems to handle aggregated traffic. I
think that is why the rates are higher for 100G in the backbones than for
40G. Considering this further, the need for 40G/100G in backbones
will exist even before deployment of these rates in the horizontal, as
aggregation is needed even when the horizontal is only deploying 10G.
1300 East Lookout Drive
Richardson, TX 75082
Alan Flatman <a_flatman@xxxxxxxxxxxxxx>
08/28/2008 07:32 AM
Please respond to
Alan Flatman <a_flatman@xxxxxxxxxxxxxx>
Re: [802.3BA] 802.3ba XR ad hoc next
Message text written by "PETRILLA,JOHN"
>In flatman_01_0108, page 11, there's a projection for 2012. There
40G, the expected adoption percentage of links in Client-to-Access (C-A)
applications of 40G is 30%, for Access-to-Distribution (A-D) links, it
30%, and for Distribution-to-Core (D-C)links it is 20%. While Flatman
not explicitly provide a relative breakout of link quantities between the
segments, C-A, A-D & D-C, perhaps one can use his
sample sizes as an estimate. This yields for C-A 250000, for A-D
for D-C 3000. Combining with the above adoption percentages yields
expected link ratio of C-A:A-D:D-C = 750:48:6.
Perhaps Alan Flatman can comment on how outrageous this appears.<
The primary purpose of flatman_01_0108 was to establish length
distributions for C-A, A-D and D-C links in the enterprise data centre.
Slide 11 records the presence of Ethernet speeds when the survey was
conducted in 2007 plus anticipated deployment for 2010 and 2012. Slide
data simply shows if any speed is present or not (it could be a single
in an installation!). It does not imply any penetration level. I included
slide 11 to observe a speed trend and it should not be used for any other
I've been watching the reflector traffic with considerable interest and
would like to clarify some of the data contained in flatman_01_0108.
99.175% of all links in this survey were under 100m. If I exclude C-A links
under 10m, and assume they are implemented in copper, then 98.8% of all
remaining links will be under 100m. I have calculated link capture rates
to 300m for both scenarios, expressed as %ages of total links. My
assumption here is that the same extended reach MMF solution would be used
from 10m up to its maximum reach. Please see the attached.
I have also calculated 40G and 100G port volumes based on hayes_01_0407
(revised in the HSE tutorial of Nov 2007), noting the contributions of
extended reach in 25m increments over time. I haven't quite finished the
graphics for this yet.
Questions or comments welcome.
Best regards, Alan
[attachment "Data Centre Links vs. lengths for XR ad hoc.pdf"
deleted by Paul F Kolesar/CommScope]