By Casimer DeCusatis
In the next few weeks, I’ll be bringing you a series of special reports from the MIT Workshop on Photonics in Servers and Routers, commonly known as “Majorca at MIT”. This invitation-only workshop is a follow-up to the IEEE Topical Meeting held five years ago in Majorca, Spain. The results of the original workshop had a significant impact towards shaping our view of the evolving photonics industry, so the organizers felt it was appropriate to revisit the central themes of this event to see how well our crystal ball predicted the future five years ago. This is an exciting opportunity to highlight some key trends in the industry, and perhaps get a sneak peek at what’s in store for future OFC technical meetings. The MIT workshop is dedicated to the memory of the late Prof. Hans Dorren, a pioneering photonics researcher and co-organizer of this workshop who passed away unexpectedly earlier this year.
The Majorca at MIT workshop brings together academic and industry leaders in photonic systems, to assess how data center optimization and computing trends are driven by photonics, and to determine system requirements for photonics (building on the results of the original Majorca conference). The workshop is technically co-sponsored by the MIT Microphotonics Center's Open Architecture System Optimization Technical Working Group, based at Lincoln Labs, IBM Research and the VICTORIES Project, a Project for Developing Innovation Systems of the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan, and IMPULSE Project of AIST. Attendees included Intel, Facebook, Ericsson, Juniper, Hitachi, Mellanox, MIT, IBM, Marist College/New York State Cloud Computing & Analytics Center, UCSD, and many others.
Topics include a microcosm of themes from OFC, including a description of the context and trends for networking and photonics in modern systems, as well as devices and system architecture. Presentations covered a wide range of topics including photonic switching and routing, quantum photonic switches, cloud computing, high performance computing, and optical interconnects for servers within and between data centers. I won’t be able to describe the entire workshop in just a few blog posts, so I’ll try to highlight some areas of interest, and apologize in advance if I skim over anything due to space constraints (if you’d like to hear more about events like this, let me know or post a comment to OFC).
All of the content for this workshop was reviewed in light of the conclusions from the original Majorca meeting regarding roadmaps for optical interconnect technology, some of which I’ve paraphrased below:
- Everything has to be made in CMOS (so making lasers is a big challenge)
- A vertically integrated company is needed to achieve optical interconnects on the same board or chip as servers…however, there doesn’t seem to be any such company left on the planet
- Single mode fiber is more scalable, and is the preferred way forward for optical networks
- Future optical interconnects (meaning those we have today) need to achieve 10X performance improvement along with a 10X cost reduction (compared with 2010 levels)
- There is a potential 10X power savings from developing optical interconnected DRAM
- Key disruptive technologies in the data center include PCM and 3D packaging
- Bandwidth bottlenecks are an opportunity, this is where the industry makes money
- Internet and data center switching should be able to solve a global problem with local information and decision making
- The Internet architecture will migrate into data centers, and the data center architecture will migrate to the board/chip level
- A key technology will be photonic “wall sockets” (a simple light source just like an electrical outlet)
This list is just as noteworthy for what it doesn’t include. For example, the term “software defined networking” (SDN) was first coined at MIT in 2009, and the fundamentals of cloud computing were articulated in the famous UCSD/Berkeley paper “Above the clouds” that same year. The original Majorca workshop doesn’t prominently mention either of these trends, which have led to new architectural and business models.
In the next blog post I will delve into the prediction made in Majorca five years ago asserting the migration of network architectures from the campus LAN into the data center, and then into boards and chips.
Posted: 18 August 2015 by
| with 0 comments