2017 Rump Session
Sub $0.25/Gbps Optics; How and When will Fiber Finally Kill Copper Cable Interconnects in the Data Center (DC)?
Chris Cole, Finisar Corp.
; Dan Kuchta, IBM TJ Watson Research Center
Andreas Bechtolsheim, Arista Networks;
Mitch Fields, Broadcom Ltd.;
Tad Hofmeister, Google;
Benny Koren, Mellanox Technologies;
Brian Kirk, Amphenol Corp.;
Ashok Krishnamoorthy, Axalume;
Beck Mason, Oclaro Inc.;
Brian Welch, Luxtera Inc.
In the DC, switch interconnects are exclusively fiber-based, but copper cables stubbornly hang on as server interconnects at ~$0.50/Gbps per link. Optical transceivers inside active cables or modules will have to be $0.25/Gbps to match cost. The high volume potential of this application can drive development of disruptive technologies and lower the cost of all DC optics. Yet our industry is moving away from common optics. Previously, focus on a few standard types like SR and LR created a huge 10G common market. Today, Operator and System OEM emphasis on optimizing their individual applications is leading to the proliferation of architectures, rates, link specifications, and form factors, fragmenting the volume of all optics types. This is accompanied by stringent port density and power requirements, making simultaneous optics low cost even more difficult to achieve, prolonging the lifespan of copper cables.
Questions for Discussion:
- What technologies are required to get to sub $0.25/Gbps optical transceivers?
- What happened to the $1/Gbps optics cost target for switch interconnects? Doesn’t industry have to hit it first before $0.25/Gbps for server interconnects?
- Are technologies for $0.25/Gbps optics and $1/Gbps optics synergistic or unrelated? Does 0dB loss budget cost less than 4dB loss budget?
- What’s the trend in switch-to-server architectures; move them apart as in end-of-row switch topologies or move them closer together as in multiple servers and switch on a card? Is there even a common switch-to-server topology to create a high volume optics market?
- Are optical transceivers inside active cables lower cost compared to inside modules? Are there operational costs that negate this?
- Does matching today’s copper cable cost even matter for connecting servers? If in a few lane rate generations, copper cables cannot support useful reaches, won’t DC Operators just have to pay more?
- How does the industry get to high volume for any optics type with the current trend of fragmentation of requirements and applications?
- Cloud DC Operators insist on low cost and bleeding edge performance on day one. If they have to make a choice, will they stay at lower cost for lower rate or pay higher cost to move to higher rate?
- Cloud DC Operators have identified power as their most critical requirement yet are restricting technical solutions, for example to SMF only. Are low-power technical solutions like VCSELs being arbitrarily excluded?
- Is Silicon Photonics the answer to sub $0.25/Gbps or $1/Gbps optics?
- Short introductory presentations by session organizers.
- One slide presentations from diverse group of industry provocateurs.
- Vigorous audience participation after each presentation, with organizers facilitating open discussion.
- Attendees come prepared with tough questions and insightful comments.