The Router That Never Went Optical

By the late 1990s, a single strand of glass could carry more traffic than the entire telephone backbone of a decade earlier. Wavelength multiplexing had turned fiber into something close to miraculous, and the industry drew the obvious conclusion: if light was this good at carrying data, the whole network should become light. They called it the all-optical network, and for a few years it was the most confident phrase in telecommunications.

There was one holdout in the picture. At every junction of the network sat a router, and inside every router the miracle stopped: light was converted to electrons, queued in memory, inspected, judged, and then converted back to light to continue its journey. The most expensive, power-hungry, latency-adding step in a photon’s life was the moment somebody had to make a decision about it. The natural dream — pursued seriously, by serious people, with serious money — was optical packet switching: let the packet stay light from one end of the world to the other, decisions included.

It is worth being precise about what a router actually does, because the dream died of that precision. A router does three things. It reads a packet and decides where it goes. It resolves contention — when two packets want the same output at the same moment, someone must wait. And it forwards. Optics conquered the third job absolutely; no electron has ever moved a bit across an ocean as cheaply as a photon. But the first two jobs concealed a common dependency, and the dependency was memory.

The only memory light has ever had: a longer road.

Light does not stop. It cannot be parked, held, or asked to wait; the only optical buffer anyone has ever built is a delay line — a longer piece of fiber, a detour. A kilometer of coiled glass buys you five microseconds, and at the end of it the packet arrives whether you are ready or not. You cannot address it, revisit it, or reorder it. An electronic router of the era could hold millions of packets in RAM and release each one at exactly the right instant; a rack of fiber spools could awkwardly juggle a few dozen, at enormous size and cost. Reading the packet’s header was no better — that requires logic, and optical logic never escaped the laboratory: a handful of demonstration gates against the billions that a routing decision was assumed to need. And contention, the third job, simply reduced to the first problem again — waiting is memory. Every corridor of the research program led to the same wall.

Around 2008, Rodney Tucker — one of the field’s most respected figures — published what amounted to the autopsy, comparing the energy and footprint of optical buffering against electronic memory and concluding, with careful arithmetic, that the contest was not close and would not become close. The field listened. Funding moved on. By 2010 the abandonment was quiet and nearly total, and it deserves to be said plainly: this was not a failure of nerve. It was a considered, evidence-based retreat by people who had done the work.

What survived the retreat is genuinely impressive. The industry kept optics for what optics does best and rebuilt the rest around it — and in the last few years, optical circuit switching has arrived in earnest: mirrors and wavelengths reconfigured in software, photonic fabrics stitching accelerators together, entire datacenter cores going electrically silent. But notice the shape of the victory. Circuit switching removes the per-packet decision rather than solving it. An electronic brain schedules the circuits in advance; the light flows through choices already made. The decision was not conquered. It was evacuated.

So the ledger, twenty-five years on, reads like this: light carries effectively all of the world’s data, and it has never once been allowed to decide where the data goes. Every routing decision on earth is still made in electrons, behind a queue, in a buffer, by logic — because the field concluded, reasonably, on the evidence it had, that a decision is a computation, and computation needs memory and gates, and light offers neither.

Every retreat rests on a premise, and that one has now stood unexamined for a generation. Whether it is a law of physics or only a habit of engineering is, as far as we can tell, still an open question.

That is the space we work in. More soon.