By Patrick McLaughlin
Singlemode fiber-optic cabling systems, once found almost exclusively in service providers’ long-haul networks or in expansive campuses, are now commonly being used in data center and enterprise networks. How deep has singlemode gotten into these networks, and how much deeper are they likely to go? That’s the topic of this month’s In-Depth discussion.
We talked to Gary Bernstein, Leviton Network Solutions’ senior director of product management for fiber and data center solutions; John Bruno, Fiber Instrument Sales’ training specialist; and Robert Reid, Panduit’s senior technology manager for data center connectivity.
While the cost differences between singlemode and multimode cabling always have been measurable, the more-significant cost differences between a singlemode and multimode system have been in the optics and lasers used to generate signals. Short-wavelength, VCSEL-based optics, supported by multimode cabling, historically have been significantly cost advantageous compared to long-wavelength optics, which are supported by singlemode cabling. In recent years however, the cost delta between long-wave and short-wave optics has been shrinking. We asked our contributors the following.
Q:Comparing today’s pricing versus that of 5 years ago and 10 years ago, can you characterize the extent to which the cost delta between long-wave and short-wave optics has shrunken?
Bernstein:As with most data center networking trends today, the hyperscale and cloud data center operators tend to be the ones driving change. Whether talking about density, termination methods or infrastructure types, market leaders and high-profile companies offer techniques to mimic or avoid. But the sheer size of hyperscale facilities along with the lightning-fast speeds required to serve their customers have made the bandwidth and reach capabilities of singlemode the infrastructure of choice for new installations.
There was a time when singlemode transceivers were typically 7 to 8 times the cost of multimode. But the purchase volumes of hyperscale and cloud data center operators have made a significant impact on the reduction in cost on singlemode infrastructure. Specifically, it is as hyperscale operators are deploying 100-, 200-, and 400-Gbit/sec channels, right as large enterprises are moving up to 100 Gbits/sec—even without the longer reach requirements—that we get cost parity of multimode and singlemode channels at those higher bandwidths.
For example, when looking at the singlemode 100G-PSM4 and multimode 100G-SR4, both parallel optic variants for a 100-Gbit/sec channel, the two options have become essentially the same price for optics and cabling. PSM4 transceivers were specifically designed as a lower-cost option for at least 500 meters of reach, using an 8-fiber MPO/MTP connection. When cost is no longer a factor, singlemode becomes the clear choice over multimode. For this reason, singlemode transceivers are expected to account for 68% of the total market volume by 2022, according to research firm LightCounting.
Just as important, the price for long-reach singlemode solutions such as 100G-LR4 has dropped significantly and will continue to drop over the next several years. According to LightCounting, in 2015 100G-LR4 transceivers could run upwards of $10,000. Today they can go for $1,000 to $2,000 depending on distance requirements.
Reid:The shift in the cost, and requirements, of higher-speed singlemode optics over the last several years has been driven by the specific needs of the hyperscale entities including AWS, Microsoft, Facebook and others. The 100G PSM4 PMD has emerged as the most successful high-volume singlemode solution and is roughly on an equal unit sales volume with 100G SR4 multimode PMD solutions.
Historically, singlemode transceiver cost sits somewhere between 2x and 3x that of same-data-rate multimode transceivers. Such singlemode transceivers typically supported 10-kilometer-plus reach. Newer singlemode solutions, catering to the needs of hyperscalers, deploy lower-specification optoelectronics to yield “sweet spot” 500-meter/2-kilometer reaches. Currently volume purchases of 100G PSM4 put unit cost somewhere between 1.3x to 1.5x that of comparable multimode 100G transceivers. This fact, combined with the significantly higher per-meter cost of higher-grade multimode fiber versus singlemode fiber, indicate rough cost parity for longer-reach applications, for the transceivers plus supporting fiber-cable plant. And of course for some very large hyperscale builds, multimode fiber solutions are out of the question because of limited reach.
Bruno:The delta between multimode and singlemode equipment was as high as 10x in the past. Over the last several decades we have seen a dramatic decrease in the price of singlemode lasers, and with that, a large cost savings in the related equipment. Currently you can purchase a singlemode 1-Gbit small form pluggable (SFP) for about $2 more than the multimode version. That multimode 1-Gig SFP can provide up to 2 kilometers of distance, whereas even the weakest singlemode SFP is 10 kilometers, with longer distances—up to 80 kilometers—available. When looking at 40-Gig SFPs, the price difference is on average $5 more for singlemode. But consider, multimode shifts its distances down to 550 meters while singlemode maintains a 10-kilometer minimum.
Q:In some data centers today, long-wavelength, singlemode-based systems are supporting speeds and distances that could be accommodated by short-wave, multimode-based systems. Can you identify one or more of these applications in use today, and describe why a data center operator would find value in using singlemode as opposed to multimode systems?
Bernstein:In 2016 Microsoft Azure, a market leader in cloud services, moved the vast majority of its data center fiber cabling to singlemode. In fact, Microsoft is now 99% singlemode, using parallel singlemode with MTP connections more than any other fiber type. Also, Facebook has undergone efforts to shorten their data center cable links to 500 meters or less, allowing them to deploy singlemode solutions at lower costs. These singlemode installations can support higher data rates and higher cabling densities at distances beyond that of multimode.
One other big draw of singlemode is longevity; there are simply fewer generations of fiber to deal with. If you installed OS1a or OS2 singlemode years ago, you would be able to support a current-generation speed at the distance specified by the standards. For example, the OS2 cable you installed 10 years ago could support a new 100-Gbit/sec network, such as 100GBase-DR at 500 meters. The connectors may need replacing, but you would not need to pull new cable. With multimode, an OM1 or OM2 would not be able to support a new 100GBase-SR4 network, OM3 could only support SR4 at 70 meters, and OM4/OM5 at 100 meters.
Another key benefit of singlemode is the ability to have more hops, or connections, in a channel. This is because the channel insertion loss budget is much higher with singlemode than multimode—around 6 dB versus 1.9 dB. This allows data center operators to have more flexibility in their network design.
Q: Looking toward the future, what standards are currently under development that ultimately will produce new applications for singlemode-based systems?
Bernstein:While next-gen network standards under development include both multimode and singlemode fiber options, the majority of 100-, 200-, and 400-Gbit/sec transceiver options recently introduced are for singlemode networks. And recent standards committee activities continue to promote more singlemode options for higher speeds.
These singlemode standards projects include IEEE P802.3cn, which will specify distances up to 40 kilometers for 50, 200, and 400 Gbits/sec. The IEEE P802.3ct Task Force will define 100 Gbits/sec and 400 Gbits/sec on a single wavelength capable of at least 80 kilometers over a dense wavelength division multiplexing (DWDM) system. And IEEE P802.3cu is defining delivery of 100 Gbits/sec (single wavelength) and 400 Gbits/sec (4-wavelength) operations for distances up to 10 kilometers.
There are also IEEE projects underway to increase the data rates of Ethernet Passive Optical Networks (EPON) to 25 Gbits/sec and higher, and distances up to 50 kilometers (Super-PON).
Q: In an enterprise environment, a passive optical LAN (POL) is a singlemode-based application that is reported to be experiencing growth in deployment. Do you believe passive optical LANs will enjoy continued growth in the years ahead?
Reid:Passive optical LAN is poised for significant growth in hospitality, high-rise commercial/office space, government and education markets. The common theme in these vertical markets is the minimization of total installed cost per outlet. Such costs for POL versus a traditional hierarchical star active Ethernet solution start to become very attractive for POL when the outlet count becomes high.
In my experience, depending on architecture, POL starts to become more cost effective at several hundred outlets per site. Power costs are also lower for sites with many endpoints. Newer POL solutions allow for centralized, closet-based low-voltage power feed to user optical network terminals (ONT) over hybrid copper/fiber cabling. This power infrastructure and hybrid cabling model is harmonized with overlay communication technologies such as distributed antenna systems—the same centralized power and cabling system can feed DAS remote antenna units.
Bernstein:Off and on, passive optical networks have been viewed as the next step in LAN, especially in K-12 and government applications. Running fiber all the way to the desk can seem like a way to clean up your network, reduce points of failure and futureproof. But in reality, the idea of running fiber out to multiple breakout points, eschewing traditional telecommunications rooms and copper horizontal cabling, has both benefits and drawbacks. It can reduce the amount of cabling required in a building, and it can have a positive effect on energy-usage reduction efforts. But it can also prove frustrating as one failure can lead to 16 or 32 users being affected, and maintenance, moves/adds/changes and repairs to the network are more complex and require more expertise—making the cost of deployment and maintenance higher.
Additionally, the energy draw and heat produced by many optical network terminals (ONT) and optical line terminals (OLT) can be deceptive, and end up requiring more space than anticipated or result in shortened lifespan of active equipment. There will still be interest in PON, along with attention to more-robust wireless systems to satisfy the mobile end user.
Q:A passive optical LAN is not the only enterprise or campus network type that includes singlemode fiber-optic cabling. Can you describe one or more enterprise-connectivity scenarios for which singlemode fiber is a legitimate, cost-effective medium?
Reid: Singlemode riser backbone cabling combined with active zones (or fiber to the enclosure) as described in the TIA Fiber Optic Technology Consortium’s LAN architecture cost model (and in the companion document on their website) and in TIA-569 (pathways and spaces) presents a cost-effective way to deliver long-reach switching to small remote workgroups and/or building automation systems fed from an equipment room or main closet without the expense and operational costs of additional local closets for these workgroups/endpoints. Multiple fiber-to-the-enclosure “zones” fed by singlemode fiber present as wall-, ceiling-, or floor-mounted enclosures containing mini-switches and small patching facilities for incoming singlemode backbone cabling and outgoing short horizontal copper zone cabling to the application.
Q:Do you have any other observations to share with us?
Bruno:I have always stated in my training classes that technically, multimode fiber really only exists for one reason. That reason has nothing to do with technology and everything to do with—you got it—money, or the expense of the network. I have always contended that singlemode fiber is always the better cabling option when considering performance, upgradability, distance and bandwidth. The real reason multimode fiber is used is because of the price of the electronics and, more specifically, the price of the light source.
But things changed for multimode several years ago with the advent of 1-Gbit/sec Ethernet. Several things happened when this new network speed hit the market. Number one was that traditional LED light sources outlived their usefulness; they could not deliver 1-Gbit/sec transmission. LEDs were replaed by VCSELs, vertical cavity surface emitting lasers, which were (and are) still fairly inexpensive. VCSELs could handle 1-Gbit/sec speeds then, and can handle 10-Gbit/sec speeds today. So, problem solved, right? Not exactly. When 1-Gbit Ethernet emerged, the networking world learned that the standard FDDI-grade fiber that was widely deployed was never tested for use with VCSEL light sources. Once they were tested, it was revealed that FDDI-grade fiber could not support 1 Gig without the use of mode-conditioning patch cords.
New grades of better-performing (laser-optimized) multimode fiber were created. Today there are five grades: OM1 through OM5, with increasing levels of performance.
Nothing stays stagnant, of course, including network speeds. VCSELs top out at 10 Gbits/sec, so when the networking world advanded to 40 and 100 Gbits/sec, the short-wavelength, multimode-based solution was parallel optics, which comprises multiple lanes of 10-Gbit transmission. For example, 40-Gig requires 4x10-Gig transmit lanes and 4x10-Gig receive lanes, for a total of 8 fibers. Likewise, 100-Gig requires 10x10-Gig transmit and 10x10-Gig receive, for a total of 20 fibers.
OM5 is the only multimode fiber that has the ability to support short-wave-division multiplexing (SWDM). In this setup, we can use four multiplexed 10-Gig signals over just two fibers (one transmit, one receive). OM5 multiplexes the 850-, 880-, 910-, and 940-nanometer operating windows. It’s getting interesting, isn’t it?
So what about singlemode? As a technology, nothing is better than singlemode fiber. But what about the huge price differences in equipment costs? As I stated earlier, the delta between multimode and singlemode equipment had been as much as 10 times. But over time we have seen a dramatic decrease in the cost of singlemode lasers and with that, a large cost savings in related equipment. So is an all-singlemode installation something we should consider? In my opinion, it isn’t something we should just consider; it is something we should encourage. One important fact concerning pricing of the fiber itself is that singlemode is always less expensive. Looking at bare, unjacketed, fiber prices, singlemode is available for $0.08 per meter, while OM3 is $0.20 per meter and OM4 is $0.40 per meter. If you calculate just that expense, for every 100 meters of singlemode cable, the fiber will cost you about $8, whereas OM3 is $20 and OM4 is $30 per 100 meters.
So how can singlemode be a more-expensive installation? When we consider the equipment (SFP) price differences between multimode and singlemode (a $2 difference for 1 Gig and $5 difference for 10 Gig), the singlemode wins every time. When considering 40- and 100-Gig applications, OM3 and OM4 require 8- and 20-fiber minimum, respectively, for multilane transmission. The fiber-price savings become staggering. Singlemode offers numerous advantages.
Patrick McLaughlin is our chief editor.
Gary Bernstein, RCDD, CDCD
Senior director of product management, fiber and data center solutions
Leviton Network Solutions
Gary has more than 20 years' experience in the communications industry, with extensive knowledge in fiber cabling infrastructure and data center architectures. Gary works closely many hyperscale companies to understand their applications and requirements. He has held positions in engineering, sales, product management, and corporate management. His data center cabling speaking engagements include AFCOM, BICSI, and Datacenter Dynamics. Gary has been a member of the TR42.7 Coper and TR42.11 Fiber Committees, and several IEEE Task Forces, including IEEE802.3bs 400G, 802.3cd 50/100/200G, 802.3cm 400G MMF and 802.3cu 100/400G SMF.
Fiber Instrument Sales
John Bruno joined FIS in 1996. As the company's Training Specialist, John is responsible for providing Fiber Optic training to hundreds of FIS customers annually. He conducts training in cities around the country, as well as at FIS facilities. John also conducts Corporate Training for groups of employees at their place of business. He holds an Associate’s Degree in Computer Science from Morrisville College and a Bachelor Of Science Degree in Computer Science from SUNY Utica/Rome. In his free time, John enjoys hockey, softball, basketball and golf. He is also a fireman with the Oriskany Falls, NY Fire Department.
Senior technology manager, data center connectivity group
Robert Reid defines product development direction for Panduit’s fiber-optic structured cabling product line. He has been active in the fiber-optic industry for more than 30 years, in the development of passive components, optoelectronic and specialty optics systems. He has presented at industry symposia and has participated in the development of the standard TIA fiber-optic test procedures. Reid also has served as membership chair of the TIA Fiber Optic Technology Consortium, and is a member of ANSI T11 Fibre Channel.