[arin-ppml] debunking the myth that Moore's law helps
Michel Py wrote:
>> Joel Jaeggli wrote:
>> I think it's awesome that the PPML lists has suddenly
>> sprouted router silicon experts.
> You're not very good at sarcasm IMHO and for the record, I never
> pretended to be one.
Which is good because your observations on the future are something like
2-5 years in the past.
>> It would be interesting to hear your projections for
>> the fib capacity of 100Gb/s ethernet centric forwarding
>> engines circa this time 2011.
> It's just a matter of time, and I'm sure when vendors feel like
> producing it I'm sure they will let you know.
If you're engaged in a silicon design exercise or capacity planning for
an isp, it's not hard to interpret the graphs on fib scaling. you've got
at least a 1st order approximation of where your FIB needs to be and when.
100Gb/s is late from the perspective of a number of operators, but it's
arriving, and in two years it will be deriguer. the silicon that's being
taped out to support it will arrive with the interfaces.
Despite assertions to the contrary stated here and elsewhere these are
not the most complex pieces of low-volume silicon ever made. Complexity
is growing, because as any observer can tell you that a some point the
benefits of going wider exceed the cost of doing so instead of going
faster. Whether moore's law does or doesn't hold, the process, speed and
density requirements are still trailing those of the most complex
components out there, which provides headroom.
On the fib scaling front, the sky hasn't fallen since the raws workshop
in 06 and it's not likely too in the scope of your planning cycles.
Operators about grouse about the capex, but so long as their internal
complexity continues to occupy a significant chunk of fib real-estate,
bitching about fib growth from deaggregation, when it's only part of the
problem is a bit strong sighted. The networks that needed 2 million
route FIBS to have reasonable headroom in 08/09 didn't need it due to
the DFZ. FIB growth is predictable enough to put in a business plan, if
your MPLS customers are paying for it so much the better.
> Reality check: GigE is on the consumer/residential market by now.
> Reality check: 10 Gigabit Ethernet has been available in some rooms for
> Reality check: as of today, 10 Gigabit Ethernet is available on all the
> market segments.
> 8-port 10GigE for the CRS-1: http://tinyurl.com/yz5hbff
> 4 and 8-port 10 GigE for the 6500/7600: http://tinyurl.com/2fu9lm
> Catalyst 4900 with two 10 GigE uplinks: http://tinyurl.com/yougjl
> Anybody in here thinks that the vendors are not working on the next step