So I now work at Google, one of the largest global data networks. Among other things this has meant I've quickly had to learn a lot about optical (DWDM) systems in the real world. Between the various ones we, and carriers we work with deploy I've had to learn the basics of a bunch of optical systems and see one big thing they're all ignoring.
Photonic switching (the high density kind, 40x40 and larger). Essentially a DLP chip (think projectors) used as a light-router. Combined with already existing, but largely unused things like commodity tunable optic modules
to allow existing routers to speak DWDM wavelengths natively.
Once combined with the giant MPLS switches about to hit the market (at least Juniper's PTX), or MPLS capable large routers (Juniper T series, Cisco CRS, Brocade NetIron) the need for very high number of individual links starts to become a real pain when having to manage even just the cabling.
Before I spell out this idea, just a little disclaimer, I know of no (announced) commercial system that is similar to this, nor am I aware of any proposed similar system being pitched to my employer or other location where I could be expected to be aware of it. This is also not some secret Google internal project (well, that I'm aware of). A quick search shows one related research attempt
, but nothing really close.
The holy grail for optical transport in IP/MPLS networks is a fully optical DWDM system, leaving all the intelligence in the routers (or a centralised control plane), while removing the need to manage the bandwidths of a line system (transceivers, multiplexers, etc.). It also allows pre-cabling everything at initial installation letting wavelenghts be brought up as needed remotely. If the router optics are tunable this even allows shifting channels with only a few seconds outage for the re-tune.
As tunable optics, filters, and photonic switches have now been in the market for several years the core components for this are ready, they just need to be combined.
The flow of my 1 degree terminal system is:Inbound
- (Optional preamp with tunable filter for long haul)
- Trunk Fibre in
- Monitor split (50% pass)
- Management channel drop
- 40/80/160 channel demux
- Photonic switch
- Patch port
- Fibre out to coloured / tunable optic on router
The multi (3+) degree wavelength router:
- Fibre in from coloured / tunable optic
- Patch port
- Tunable filter and attenuator (Band protection, power matching)
- Photonic switch (for a 1-degree site this really just buys loopback)
- Optical mixer (potentially inside the photonic switch)
- Power amp (Likely RAMAN/EFDA combination)
- Management channel merge
- Monitor split (99% pass)
- N inbound trunks & demux, same as 1-degree up to photonic switch
- N outbound trunks same as 1-degree from filter, skip switch
- M wavelength conversion transponders. These would be electronic (two tunable XFP / SFP+'s in a trivial media convertor design) but only in line when conversion actually needed. If and when pure optical wavelength conversion breaks out of the lab these could be used instead.
The photonic switch in this case needs to be much larger in this case, 200 inputs probably being the minimum to handle the expected transit plus transponder traffic.Extra bits:
- Use some spare ports on the photonic switch to allow (remotely triggered) hard loops on the line system to aid link troubleshooting
- The management port is just one locked wave (possibly 1300nm) running ethernet
- If there was any way to usably have an ODTR permanently hooked to the trunk TX/RX fibres (not permanently running, but automatic on failure of management channel) that would be ideal
- The trunk side of things could simply be a JDSU "AON Super Transport Blade" with the WSS removed
- The Glimmerglass photonic switches are the obvious candidates for switches, although they only go to 192x192 with simple design a CLOS tree should allow expansion to N-degree M-wavelength systems
If used as a metro system this could even not need the amps or attenuators, making this a purely passive system, yet fully switchable. IF you trusted the optics enough (or just accepted the occasional interference) you could drop the filters as well, reducing this to an off the shelf optimux going into an off the shelf photonic switch (although the cabling load would kill it if you tried to build it like that).
In theory a 40 wave, 1 degree terminal built against this should easily fit in 6RU or less, and cost much less then the active systems of today (albeit more then the fixed passives of today). A 4 degree interconnect should fit in 8-10RU plus any space needed for wavelength converting transponders.