Five Minutes with Brad Smith of LightCounting (Part II)

Last week, we brought you the first half of our recent conversation with Brad Smith, senior vice president of LightCounting. Below is the second installment of our discussion with him, where he raises a few key points about active optical cables that I think you’ll find interesting. Whether you agree or not, feel free to weigh in in the comments below.

Lightspeed (LS): Why are Active Optical Cables adopted in InfiniBand clusters?

Brad Smith (BS): Infiniband users are “speedfreaks” and early adopters of anything fast. The speed needed to get from point-to-point requires very short latency without a lot of protocol or signal processing overhead or line delays. Electrical signals take time to go through a copper cable and the time delay gets worse with distance (not so however with light). AOCs minimize any overhead and there is no difference in delay at 1cm or 100m!

With copper interconnects, as speed increases – everything gets worse. Connectors and cables get fatter, heavier, more costly, consume more power and the reach drops off dramatically. With optical, it’s the opposite. Infiniband managers are worried that the sheer weight of the copper cables will cause the systems to fall over turning the aisle ways into hard hat zones! With AOCs, the size, weight, reach and aggregate data rates are almost becoming non-issues! Some approaches are trying to include signal processing chips inside the connector to extend the reach. These are last ditch, desperate efforts for the copper interconnect world, but the writing is on the wall.

Infiniband systems are in a “cluster” because of the signal delay and short reach of copper interconnects. Datacenter operators would like flexibility in their floor layout and put the servers and storage where they want and not be dictated by the reach of copper cables. AOCs enable this to happen. The next speed jump to 40Gbps/line will blow out copper almost entirely as the reach will drop to a few meters. What then? Systems implemented in copper will be Infiniband “spheres”?

LS: Do you think that Active Optical Cables will be adopted in Ethernet? Where and when?

BS: In the next 2 years, the data center will upgrade to faster multi-core AMD, Intel servers and start implementing FCoE at 10Gs to support the massive workloads being developed by email, internet and server virtualization. 10G will become the standard “interconnect currency” between servers and switches and other hardware and will be used internally throughout the system internals. But the system-to-system interconnects will need multiples of 10G, meaning multiple 40G-150G interconnects. FCoE will simply push some of the Fiber Channel traffic into the 10G Ethernet space, making things more demanding. AOCs fit these requirements as the reach is typically <50-100m and high aggregate data rates are needed as inexpensively as possible. Even Twin-ax copper is simply not an alternative. Top-of-rack to end-of-rack interconnects, rack-to-rack and to core switches and routers will be where AOCs are adopted as they are an ideal solution.

40G and 100G Ethernet are “just around the corner” but no one talks about the corner of what. With enormous technology and standardization challenges still ahead, volume adoption of 40/100GbE in data centers is probably several years out. While initially adopted in Infiniband applications, we do see AOCs like Finisar ‘ s Quadwire and C.Wire cables driving early 40/100GbE pre-standard adoption for short reach interconnect uplinks in the data center.

A lot can happen in 5 minutes

We recently had the opportunity to sit down with Brad Smith, senior vice president at LightCounting, a research firm dedicated to the transceiver market. Brad oversees LightCounting’s transceiver-related semiconductor and optical markets coverage, so he has a unique view on active optical cables (AOCs). He was gracious enough to answer our questions and share his perspective with us. Part 1 of our conversation is below for your reading pleasure.

At the same time, we’re also pleased to announce the debut of the Five Minutes With… format, a recurring feature on Lightspeed, where we’ll speak with industry experts, who will provide their insights and opinions on a particular topic. We look forward to bringing you more of these in the coming months and welcome your feedback. Let us know what you think in the comments below.

Five Minutes with Brad Smith of LightCounting

Lightspeed (LS): When was the first time you remember hearing the term ‘active optical cable’?

Brad Smith (BS): In a 2007 Lightwave magazine article which showed a picture of Finisar’s Laserwire and its AOC connector-end. I thought it was an optical USB connector.

LS: In the early days, what was your market prognosis for active optical cables and, in hindsight, are you surprised about where the industry is now?

BS: IBM is the technical high ground for supercomputers and they said, (I’m paraphrasing here) “We are done with copper and going optical interconnects from now on.” Clearly AOCs are a great fit for Infiniband and HPCs. But the bigger market is in the corporate data center. With the exponential increase in data and traffic, coupled with the next AMD, Intel server upgrade, the data centers are clearly feeling pain and seeking faster interconnects beyond 1G. What is surprising is how difficult it has become to implement 10G in copper and the RJ-45 jack. The walk up from 10, 100, 1G was relatively non-eventful and the reach was always 50-100m. Implementing 10G in copper is turning out to be very hard and is still roughly two years away, whereas 10G, 40G and 120G AOC interconnects are available today and can alleviate a bunch of problems!
With AOCs, by simply closing off the optics to the end user, a huge number of limiting factors for optical interconnect adoption disappeared and innovation has popped with a large drop in cost. This reduced cost coupled with the large number of interconnects typically required in a data center is enough to gain considerable attention by purchasers in many applications. LightCounting is forecasting that the AOC concept will spread outside Infiniband to Ethernet, Fiber Channel, SAS and other areas that need 10Gbps and reach beyond 5m.

LS: What do you believe are the key advantages of AOCs?

BS: In my view, the main advantages are:

• 10G Reach – As speed goes up, copper reach goes down. With AOCs, 1cm or 100m makes no difference! 10G is available now with AOCs and data center managers don’t have to wait 2 years for 10GBase-T solutions. Data center architects want to put the systems where they want and not be limited by interconnect reach issues.
• Aggregate data rate – 10,40, 120, 150Gs – With a single 12x AOC, systems architects can have a short reach 100G channel today and they don’t have to wait for the single-fiber, 100G IEEE standards to sort things out. Also, a system designer can implement 1Tb over 100m+ with just 7-8 12-channel AOCs (12x10G, 12×12.5G). A few years ago, 1Tb was the rate the industry used to describe the aggregate data traffic between countries! Finisar’s 12×12.5G AOCs puts 150 1G links into the diameter of one Cat-5e cable.
• Low Power consumption – Data centers consume so much power that collectively it is being measured as a percentage of a nation’s electricity consumption figures! Large datacenters can consume $1-2M/month in electricity to power and cool the electronics. Today, it is becoming a major limitation both technically and financially. A single 10GBase-T copper interconnect today consumes ~10-15W per line card end and promises 6-8W with new chips in a few years. Finisar and a few other vendors’ 12x10G AOC consumes 1-3W/end and that is for twelve 10G channels, not just one!
• Other considerations – Size and weight of the cable, simplicity of design (compared to 10GBase-T chip complexity) and the price convergence with copper. When one examines the total cost of ownership (TCO) including power, maintenance and related issues, AOCs clearly make a lot of sense. At the TCO level, AOCs are on par, if not better than copper interconnects.

Stay tuned to Lightspeed for Part II of our conversation with Brad coming soon.