Yes, optical packet switching (OPS). The way i understood it, OPS, at least per NTT, transmits each IP packet in parallel over a number of wavelengths, with the packet header being constantly on select 'control channel' wavelengths. Only those control channel lambdas needs to be processed (AFAIK, still electronically) to determine to outbound fiber and wavelength group/band of each given packet.
There's innovative approach there certainly. From practical perspective though, if i understood NTT OPS correctly, the problematic issues that remain include:
- The lower layer processing (WDM) in OPS makes assumptions about its client layers, eg header length, packet length (ie how many wavelength timeslots the pkt header and payload can consume). Good server layer protocols, such as traditional circuit-switching WDM, make no assumptions about their client layers. Ideally, server layers should accommodate, and even adapt to, any client layer services, instead of having to make the client layer applications to fit the server layer capabilities (eg header length, packet spacing etc).
- How to keep the byte 'columns' in sync across number of wavelengths, over the existing fiber/WDM networks?
- OPS is still hop-by-hop *packet switching*, suffering from all the queue management, jitter and packet loss QoS and complexity problems that we have with electrical packet switching. (I think the Stanford's clean slate approach is on the right track here -- http://cleanslate.stanford.edu....
Would still appreciate any links to recent developments, by UCSB etc, that could lead to breakthroughs in the Internet infra space. For such breakthroughs to actually materialize in the commercial networks, they need to address both the needs for innovation (new revenue generating service capabilities, or transcending TCP/IP's built-in limitations, eg per Pouzin Soc's RINA approach) and practical viability (bottomline operational cost-efficiency).
NTT did some interesting stuff with all optical routing, as has UCSB. That's pretty out-there.