I read with interest a recent article in the San Jose Mercury News (Live Video) about how the availability of better quality cameras on smartphones and the growing appetite for on-demand content on social media now have Facebook and Twitter competing head to head to encourage more people to stream raw footage. Pre-recorded videos on web pages are the norm now but the growth of live videos is exploding. Twitter recently purchased live-video-streaming app Periscope which already has over 200 million broadcasts. Add to this the recent release of Snapchat’s Chat 2.0 which emulates face-to-face communication while making it easier to switch between video, texting and calling. Upon reading all of this I realized that it will be people like my 17 year old daughter who seem to be forever glued to their smartphone that will push the use of silicon photonics ICs (PICs) into the mainstream. What, you ask do 17 year olds and live video streaming have to do with PICs? Just imagine how much bandwidth will be required within and between our data centers and mobile devices to handle random, bidirectional, live video streaming for Facebook’s 1.6 billion users. It boggles the mind. Layer on top of this the latest ideas about the Internet of Everything where millions of smart devices will be vying for your attention and you will quickly realize that the current 10Gbps connections in today’s data centers will be woefully inadequate to handle the amount of traffic that is coming their way.
Data Center providers are already responding to the increased traffic needs by building mega data centers
(Mega-Datacenters are the Future) but these centers come with their own new challenges. Traditional data centers dealt mainly with independent jobs with only course inter-server interactions. Memory and storage were co-located close to the servers meaning that latency across the data center was not a big concern. This is changing rapidly now with the advent of software configurable mega data centers that use virtualization of processors, memory, long term storage and networks to provide unique and customized services to their customers. Virtualization brings with it the need for scalable lower latency network communications across the data center as decisions about how resources will be combined are held to the last moment and change over time. The new mega data centers are now looking at network requirements across the data center that are much more akin to what would be seen in high performance computing environments using fine grained inter-process communications. Add to this the sheer size of the mega data centers stretching to kilometers and copper wire networking gets very expensive both in terms of latency (including multiple hops through switches to get across the data center) as well as power consumed to drive those switches and cables.
Enter the advent of Silicon Photonics and new network architectures that use modulated laser light through fiber connections to reach across the 2km data center in a fraction of the time required for copper cables and switches and at a fraction of the power. Silicon Photonics will not only enable faster, less power consuming communication but it will also enable more direct communications by integrating the photonics either on die or in-package with the server’s components eliminating the need for multiple levels of power hungry switches. A good example of commercial progress towards this end is the recent product announcement by MACOM for 100G silicon photonics-based communication solutions within the data center using standard QSFP28 optical connectors (MACOM).
Nothing drives invention and paradigm shifts like necessity and I can’t think of too many forces of necessity stronger than my 17 year old daughter needing to communicate with her friends, unless it would be my wife needing me to take out the garbage.Share this post via: