By Patrick McLaughlin
On Thursday, June 6, we at Cabling Installation & Maintenance hosted an online seminar titled “Edge Data Centers.” The three sponsoring companies, Chtasworth Products Inc., Panduit, and Rittal, also provided technical and market experts who served as the seminar’s speakers.
This article is a summary of some of the information that Robert Reid, senior technology manager for Panduit’s data center connectivity group, delivered during that seminar. As author of this article, I have shaped it to be my best technical interpretation of Reid’s presentation. Some of this article comes from a transcript of his remarks, and other parts of the article are paraphrased from those remarks.
Here, then, is a look at what the industry is calling CORD, based on my reporting of Robert Reid’s presentation from June 6.
In the midst of edge computing and the growing role of edge data centers, a practice has emerged that frequently goes by its acronym, CORD. CORD stands for central office rearchitected as a data center. In a CORD deployment, a service provider uses an existing facility that has served as a central office. In some cases, these facilities now serve as headends for the service providers’ microwave towers or land lines.
Many of these buildings were built decades ago, and as the service providers’ networks have evolved, so has the central office’s functions and the equipment within it. Often, the central office has been largely hollowed out, and the footprint of equipment inside the facility is minimal. Yet these facilities still have carrier-network feeds coming into the building, and often, technologies including wireless and Gigabit PON (GPON) coming out. This combination of available space and incoming/outgoing connectivity makes the central office an ideal site for a data center.
For a service provider, the facility represents a significant capital expense and operational expense—especially if they are compared to the capex and opex of an over-the-top (OTT) content provider like Apple or Netflix. AT&T alone operates 4000 to 5000 central offices. Because these facilities originally were built as central offices for telephone services, they are by design close in proximity to residential and business users. Depending on population density, a central office may serve from 10,000 to 100,000 subscribers.
Objectives and architectures
With CORD, the objective is to build a low-latency network inside the central office. Typically a leaf-spine network architecture is a key element of that low-latency performance, and leaf-spine is prevalent in CORD facilities. AT&T, with multiple thousand central offices, is driving the CORD initiative along with the Open Networking Forum (ONF). A specific goal of CORD is to replace today’s purpose-built hardware devices with their more agile software-based counterparts and make the CO an integral part of every telecommunications provider’s larger cloud strategy, enabling more-attractive networking services. In doing so, CORD unifies the following threads: software-defined networking, network-function virtualization, and cloud services.
It has been established in cloud-scale data centers that virtualization reduces capex. That’s true with CORD facilities as well. One way to look at it is that CORD leverages cloud capability with virtualized commodity platforms and scaling, scaled down to a smaller version.
As is the case in cloud data centers, in the smaller-scale CORD facility, fiber-optic cabling solutions support a leaf-spine architecture. By doing so, they enable low-latency computing for edge services, particularly including triple play.
Looking at it from a network-architecture standpoint, a CORD data center has a switch fabric that basically is a microcosm of what happens in the cloud. The leaf-spine architecture eliminates the hierarchical switched network, in which a packet must climb all the way up through the cloud, then back down again. Leaf-spine facilitates local switching with no more than a single hop to get from switch to switch. That is critical to minimizing latency.
In addition to a leaf-spine architecture, GPON can be a vital element of a CORD facility. Today GPON is widely deployed in municipalities, as an access technology for triple-play services. The optical line terminals (OLTs) in a GPON system commoditize connectivity to the access network. Along with other access technologies, GPON plays a critical role in building a CORD network to serve end users.
In serving residential and small-business users today, a service provider is likely to use GPON to deliver a 2.5-Gbit/sec downstream and 1-Gbit/sec upstream asymmetric connection, which is fed by 10- or 40-Gbit/sec Ethernet uplinks to the service provider. While that spine-leaf fabric inside the CORD may be 10- or 40-Gbits/sec today, these service providers must plan for the infrastructure they will need down the road to serve their customers when OLTs are upgraded. Emerging technologies like XGPON, 40GPON, and 100GPON will serve higher speeds to end customers, requiring an upgrade to the infrastructure inside the CORD as well.
Fiber options
We can look to cloud providers for examples of leaf-spine architectures that are running at 40- and 100-Gbits/sec today. 400G transceivers are now reality, being sold by at least two original equipment manufacturers. For cloud providers, the transition to 400G can happen very quickly, and those speed increases will cascade all the way to users. When that happens, CORD facilities will be part of the upgrade cycle. Those that are operating at 10G or 40G today will need fiber-optic cabling media that will service 200G, 400G, and maybe even 800G going forward.
Because the scale of a CORD is smaller than a standard on-premise data center or a smaller cloud data center, multimode fiber is a good solution for the leaf-spine interconnect. Research conducted by the Ethernet Alliance in 2016 indicated that, excluding hyperscale data center facilities, a 100-meter reach will support 90% of channels, while a 150-meter reach will support 96.5% of channels. The typical channel length in an on-premise data center is about 30 to 40 meters.
If a CORD or other data center facility invests in a leaf-spine architecture supported by a multimode fiber-optic cabling infrastructure, it is in the facility’s best interest to know that the infrastructure will be able to service future speeds and in particular, future transceiver technologies. Fortunately, a technology roadmap is in place for 4-, 8-, and 16-fiber short-wavelength (multimode-based) transceivers to speeds up to 200 and 400 Gbits/sec. The 4- and 8-fiber solutions incorporate wavelength division multiplexing, which enables the operation of multiple wavelengths over a single fiber. The 16-fiber solution implements a single wavelength over each fiber.
Ultimately, users will decide which construction is optimal for them. But regardless of the choice they make, service providers that rearchitect their central offices as data centers will have viable multimode-fiber cabling systems and short-wavelength transceiver options to support low-latency, spine-leaf architectures now and in the future.
Patrick McLaughlin is our chief editor.
The ONF, CORD, SEBA and more acronyms
“On its southbound side, VOLTHA communicates with PON hardware devices using vendor-specific protocols through OLT and optical network unit adapters.”
On the edge
We make our web seminars available for on-demand viewing for 6 months. If you'd like to see and hear from these industry experts, visit cablinginstall.com/webcasts. You'll find this seminar, along with our other on-demand and upcoming events.