Will Photonic Computing Save the Datacenter?

Earlier this week I was reading an article in the Technology Review describing the first complete photonic communication system developed by chip manufacturer Intel's photonics lab. This development will lead to some tectonic changes in enterprise computing.

Instead of using copper or other electrical conductors to transmit electrical signals to the processors, Intel developed technology that can encode and decode optical signals natively on the processor, which means there is no more performance hit converting optical signals into electrical impulses or vice versa.

The immediate benefit is a serious performance boost. According to the article, the four-laser system the Intel lab staff demonstrated has the capability to carry data at a rate of 50 gigabytes per second compared to the 10 gigabytes per second achievable over copper wiring. This performance is also scalable depending on how many lasers with varying wavelengths are used to transmit data. The article's author reports that a system could handle up to 1,000 gigabytes per second.

The second benefit, which I think is a little more exciting, is the fact that the computer is no longer tethered to the physics associated with copper wiring. The system memory could be housed separately from the processor, maybe a foot away, according to Intel officials. By doing this, not only will it reduce datacenter cooling costs, but it is likely to be the last step in completely deconstructing the concept of the beige box server that we know and love.

For the past several years, the industry has gone from individual rack-mountable servers to blade servers housed in a common chassis. Now with this ability to separate memory from the processor, the datacenter becomes the chassis. Instead of having individual servers, the industry will see processing fabrics interfacing with memory fabrics and connecting with storage fabrics.

Of course, this is quite a ways down the road, since the enabling technology is still in its nascent stages. However, it will be interesting to watch how fast the financial services industry will embrace the new technology once it is available. Considering the stark performance difference between the current and new technologies, I'm not sure if the typical adoption curve will apply.