[LINK] terrabit per second internet
pbrooks-link at layer10.com.au
Wed Sep 9 13:09:57 AEST 2009
David Lochrin wrote:
> On Wednesday 09 September 2009 11:41, Richard Chirgwin wrote:
>> As you say, you'll need ten fibres (for example) to run a terabit Sy-Me;
>> but the passives that CUDOS is working on (not only them, there are
>> plenty of other researchers in this particular game) would, for eg,
>> split / combine the terabit "trunk" across the multiple bits of glass
>> without electrical conversion.
> Interesting, however I wonder whether splitting & combining data streams is worth the effort? The complexity and cost of splitting & combining would be high at the speeds involved, and what would it achieve? There's probably no single type of data stream which requires a data rate faster than can be provided by one fibre, so the best cost/benefit might be obtained by using multiple fibres to provide asynchronous capacity for independent bit streams.
Possibly not in a point-to-point transmission application, but the value
of the high-speed switching would be in the heart of a next-generation
core router, where the incoming datastreams from many different
interfaces are all re-routed and re-directed to the intended output
interface - the physical distances are very short inside the device, and
the projected traffic that a future core router might need to handle are
very large - the new technology would allow such a router to work with
many many optical fibre interfaces.
> I don't know if this actually occurs, by I imagine small changes in the group velocity of photon bundles in individual fibres, perhaps due to mechanical or temperature stresses, could play havoc with splitting & combining at those data rates.
> Or is this too naive?
Nope, not naive at all - those 'small changes' already occur between
adjacent DWDM channels simply through differential transmission speed
due to chromatic dispersion - refractive index is different for each
frequency, so different colours travel at different speeds. The current
100Gbps standards for long-haul applications currently avoid this by
using two channels at the same frequency, but opposite polarisation - so
chromatic dispersion doesn't apply, but polarisation-mode dispersion
will (fortunately PMD is usually much lower than CD). When 40Gbps
signals were being synthesised using 4 x 10G signals at different DWDM
frequencies, combining the four channels at the other end of a long
fibre did require variable timing compensation for the slightly
different speeds of each colour - no different to the timing buffer
required in any other inverse muxing situation, but operating at optical
rates - which is quite tricky to achieve.
More information about the Link