Fiber Optics For Hyper-Convergence—Make Sure Your Glass Is Not Half Full
Who can argue with the promise hyper-convergence delivers? Combining storage, compute, and networking into a single system, built for scalability, and using x86 servers—it is an intelligent design to increase complexity and increase scalability. These new facilities are designed for data-intensive and mission-critical workloads, just waiting for 5G and IoT data transferred from edge networks. According to International Data Corporation (IDC), Worldwide Quarterly Converged Systems Tracker, worldwide converged systems market revenue increased 1.1% year over year to $4.2 billion during the fourth quarter of 2019 (4Q19).
The DNA of hyper-converged data centers
Hyper-converged data centers perform differently so they must be designed differently. This means out with the old three-layer topologies that have bottleneck issues from oversubscribed uplinks and in with the new leaf-spine design. The leaf-spine design is a fully meshed environment that has many connections to each switch that limit bottlenecks and boost scalability. Transferring data through the leaves are fiber line cards connecting trunks and then going up to coupler panels.
Even though the leaf-spine design is an innovative approach for processing data, workloads are not all created equal, nor are the fiber optics that carry the signals. The 2019 optical fiber Communications Conference held in San Diego, California, underscored the importance of selecting the right fiber connections, as bandwidth was the hottest topic. During this event, it was referenced how new services such as 5G and traditional cable companies are bringing fiber links closer to their end-users and the increased need for seamless performance between multiple data center buildings. All of these musings were predicated on adapting software-defined networks—it’s no coincidence this is also the very foundation of hyper-converged data centers.
Redesigning data centers into new hyper-converged environments gives engineers and facility managers the opportunity to rethink their cabling plant. On the table for debate are the different media types which include Twinax Direct Attach Copper (DAC) and Active Optical Cables (AOC) as well as the associated transceivers, patch cords, and jumpers. Lessons learned from DAC deployments involve the standard length that often causes cable congestion as opposed to AOCs that can run longer distances with precise lengths to make connections, thus reducing service loops. Also giving the edge to AOCs is the fact that fiber jumpers in harnesses are forward compatible to handle the ever-increasing speeds that will be required as workload intensity increases.
Speeds and Feeds
Even with the transition from copper to fiber, data loss should still remain top-of-mind. Fiber loss reduction has always been difficult because the loss is so close to the fundamental limit. The Shannon–Hartley theorem tells the maximum rate at which information can be transmitted over a communications channel of a specified bandwidth in the presence of noise. As bit rates continue to increase, the industry is approaching Shannon’s limit and pushing ever closer to the bandwidth wall for a given channel.
The software-defined, hyper-converged scalability is not worth it if the data transmissions are experiencing significant latency—that’s not much of an ROI! To address this more fibers and cross-connections between active equipment, traversing ever-increasing distances will be needed, especially as data speeds migrate above 100Gb, on to 400Gb transceivers that are more powerful than their predecessors. Yes, data center managers are still using copper cabling from server-to-switch in the rack, but as higher speeds become the norm, this solution will prove to be insufficient, and more fiber optics, especially to the servers, will have to be deployed.
Using more fiber causes the back-office accountants angst because it’s more expensive than using copper. However, more data centers are beginning to utilize new Lite transceivers, which are capable of consuming 30 percent less electricity or approximately 1 or more watts saved per transceiver. Facility managers know the dramatic ripple effect that lower electrical consumption has on cooling, in fact, it has been found that for every 1 watt saved at the device level that 3-5 watts can be saved at the facility level when cooling is taken into account. Watts saved are like pennies to compounded interest, e.g., lowered cooling temperatures enabled by Lite transceivers can save $50,000 per year in smaller data centers, and around $1,000,000 per year in hyperscale data centers.
Thoughtful Fiber Migration Is Needed
When migrating to higher speeds, latency generated by bit error rate is a typical occurrence, because the transceivers have a much smaller window to hit. Hyper-converged data center managers looking to up their game to 400Gb should consider something like advanced fiber systems that have the lowest documented total channel connection loss. For example, the industry-standard fiber channel connection lost budget is 2.3 dB. By contrast, ultra-low loss fiber channel connection lost budget is 1.7 dB. Acquiring any fiber optic connectivity solution under 1.7 dB will give any hyper-converged data center a particular advantage. Note that the total channel loss of light power from point A to point B must always be calculated by measuring from patch cord-to-patch cord and cassette-to-cassette for a true analysis.
A precedent has been set; low-loss fiber optics have revolutionized the telecommunication industry for almost five decades—since the first low-loss optical fiber with less than 20 dB/km at 632.8 nm was deployed in 1970. Or put another way, we have known that electrical signals have difficulty traveling beyond the box and turned to fiber since Donna Summer was crowned the Queen of Disco. However, as outdated as disco has become, so too is the notion that optical signaling requires costly cables and connectors with no return on investment.
Hyper-converged data centers will be upgrading their cable plant to embrace more and more fiber optics on their way to delivering lower latency demanded by 5G, IoT and a host of other new data-generating sources. If careful consideration is not used when selecting the fiber and associated connections for designs, then latency will become a burden. When this happens, a hyper-converged data center’s glass is truly—half full.
Did you like this byline? Contact BVM and let us write a product story for you and place it in the press.