When it comes to choosing optical fiber systems for the data center, where and what you buy can impact your bottom line. While cable assemblies from inexpensive unfamiliar vendors such as generic cable assembly houses have their appeal from a capital expenditure standpoint, it is important to understand end face geometry, mechanical reliability, and performance specifications so you know what to ask and can ensure you are getting a product that will not ultimately cost you more in the long run —especially considering that the average cost per downtime incident can be upwards of $200K, or even more in the financial sector. And when it comes to choosing your type of fiber, that choice too can ultimately impact your bottom line. However, it can be difficult to make sense of all the choices — multimode OM4 and OM5, singlemode OS1 and OS2, and more. Let’s take a look at the key factors involved in ensuring cost-effective optical fiber systems for the data center, including reliability, performance, scalability, and application assurance.


Article Index:


TAKE IT AT FACE VALUE

The end face geometry of optical fiber connectors is an essential characteristic of repeatable and reliable connections in the data center, as well as overall transmission performance of the system. Whether deploying duplex optical fiber connectors such as LC type connectors or multi-fiber push-on (MPO) connectors (i.e., MTP connectors) used in parallel optic applications like 40 Gigabits per second (Gb/s) (i.e., 40GBASE-SR4) and 100 Gb/s (i.e., 100GBASE-SR4), performance and reliability depends on the mechanical characteristics that control alignment and physical contact of the fiber cores.

Figure 1: Simplex and duplex optical fiber end face geometry parameters

Figure 1: Simplex and duplex optical fiber end face geometry parameters.

While some optical fiber cable assemblies (e.g., jumpers, trunks, and hybrid equipment cords) from generic sources may carry a lower price tag, the overall end face quality and geometric parameters may not be in compliance with industry standards. These end face geometry parameters differ between simplex or duplex optical fiber connectors and larger array MPO connectors that need to maintain consistent fiber protrusion height across multiple fibers in the connector.

Critical end face geometry parameters for simplex and duplex optical fiber connectors as defined by IEC PAS 61755-3-31 and shown in Figure 1 include:

  • Radius of curvature (ROC) is the roundedness of the ferrule’s end face surface. A smaller value means a smaller contact area, which can ultimately increase deformation of the glass.

  • Apex Offset is the degree to which the dome of the end face is centered. It is important to minimize Apex Offset to achieve a good core-to-core contact when mating connectors.

  • Fiber protrusion height is the distance that the fiber core extends out of or recesses into the ferrule. While some fiber protrusion is required for good core-to-core contact, too much protrusion can impact overall durability of the fiber. In contrast, a fiber that is recessed into the ferrule can exhibit poor insertion loss performance or deformation of the ferrule.

Critical end face geometry parameters for MPO optical fiber connectors as defined by IEC PAS 61755-3-31 and shown in Figure 2 include the following additional parameters that take into account the differences between the individual fibers of the connector.

  • Angle of the polish on the horizontal (X) axis (RX and GX) and on the vertical (Y) axis (RY and GY) for each fiber is critical to ensuring consistency across all fiber and for maintaining proper alignment when mating connectors.

  • Fiber protrusion height (H) for each fiber is the distance that the fiber core extends out of or recesses into the ferrule as previously explained for simplex and duplex connectors.

  • Maximum fiber height differential among all fibers (HA) is the maximum height difference that can occur among all fibers of the connector (i.e., difference between the shortest and the tallest fiber in the array). When differences in fiber height exceed the standard, some of the fiber in the connector may exhibit higher insertion loss due to gaps when mating.

  • Maximum adjacent fiber height differential (HB) is the limit in height difference that can occur between any two adjacent fibers.

Unfortunately, not all vendors of optical fiber components subject connectors to end face geometry analysis or utilize stringent polishing techniques to ensure that connectors meet these parameters. The result of noncompliance is often an increase in insertion loss and return loss performance, which can ultimately impact transmission, especially in advanced 40GBASE-SR4 and 100GBASE-SR4 applications that have more stringent insertion loss requirements.

In addition to end face geometry, surface defects (e.g., chips, scratches, and pits) and cleanliness (i.e., loose debris) can also impact optical performance. Even the smallest speck of debris can cause increased insertion and return loss. Because defects and cleanliness are not always detected during end face geometry analysis, it is important to also clean and inspect every connector per the IEC 61399-3-35 standard. This standard includes certification criteria for defects and cleanliness based on connector types and fiber sizes, and inspection devices are available that automatically test to this standard. While contamination can often be removed during the fiber cleaning process, scratches, and defects may remain on connectors from vendors that do not have stringent polishing and inspection processes in place. In fact, in a recent independent testing of LC and MPO connectors, approximately 25% of the samples from generic assembly houses failed inspection even after proper cleaning.

MPO fiber end face geometry parameters

Figure 2: MPO fiber end face geometry parameters (four fibers shown for clarity) Source: IEC PAS 61755-3-31

Top

 

UNDERSTAND TELLTALE PERFORMANCE SPECS

Insertion loss is commonly used as the basis for acceptance testing of installed optical fiber links and channels. Although return loss testing of installed cabling is not required by industry standards, it is a normative requirement for optical fiber connectors and assemblies. Return loss is critical to optical performance of links and channels because reflected optical signals can interfere with detecting the optical signal by degrading the signal to noise ratio. For these reasons, it is essential to ensure that optical fiber cables, components, and assemblies are fully standards compliant for return loss and insertion loss per IEC-61755-3-31 and TIA 568.3-C standards. Prudent vendors factory test their optical fiber assemblies in both directions and at each wavelength designated for operation (e.g., 850nm and 1300nm wavelengths for multimode and 1310nm and 1550nm wavelengths for singlemode) and provide the performance test results with each component, typically via a traceable serial number.

It is important to note that while many optical fiber connectors will meet the ISO/IEC and TIA insertion loss specification of 0.75dB, most industry professionals believe this specification to be an outdated value as most MPO and LC connectors are in the range of 0.2dB to 0.5dB. Furthermore, an insertion loss of 0.75dB would eliminate the ability to use cross connects that offer flexibility, manageability, and easier deployment in 40GBASE-SR4 and 100GBASE-SR4 applications where the maximum channel loss is only 1.9dB for OM3 and 1.5dB for OM4. For example, deploying a cross connect at the main distribution area (MDA) and at intermediate distribution area (IDA) within an OM4 40 Gb/s or 100 Gb/s channel would require four connectors, but connectors with a 0.5dB performance would allow for only three connectors with no additional headroom. There is therefore a movement within standards organizations to reduce the insertion loss specification from 0.75 to 0.5dB or lower.

In recent testing of MPO trunk assemblies, the majority of samples from generic sources exhibited insertion loss values greater than 0.50dB, with some actually failing the 0.75dB insertion loss specification. Insertion loss and return loss performance are also directly related to end face geometry and cleanliness. In fact, all of the samples that failed insertion loss and return loss testing also failed to meet the fiber height parameters in MPO end face geometry testing.

Top

 

KNOW THAT MECHANICS MATTER

Mechanical reliability of optical fiber assemblies is important to ensure that these components can endure the various conditions that may arise in a real-world installation, and that they can dependably withstand the internal stresses imposed by spring-loaded physical contact over time in a variety of environmental conditions. There are several Fiber Optic Test Procedures (FOTPs) required as part of TIA-568-C.3 industry standard specification for mechanical reliability, including Flex Testing, Torsion Testing, Pull Testing, Cable Retention, Impact Testing, Vibration Testing, Durability, and Transmission with an Applied Load. Specifications for these FOTPs include:

  • FOTP-6: Cable Pull, 50 Newtons @ 0 degrees for 5 seconds
  • FOTP-1: Cable Flex, 4.9 Newtons, 100 cycles
  • FOTP-36: Cable Torsion, 15 Newtons, 10 cycles
  • FOTP-6: Cable Retention, 19.4 Newtons @ 90 degrees for 5 seconds

In the recent testing of LC and MPO assemblies, all generic samples subjected to mechanical stress tests ultimately failed mechanical reliability specifications. Some of these samples could not be tested for every specification because they failed Cable Pull or Cable Flex testing and could therefore not be subsequently tested for Cable Torsion or Cable Retention. Figure 3 shows some examples of generic optical fiber assemblies that failed Cable Pull and Cable Flex testing.

 

Figure 3: Generic MPO assembly after cable flex test and LC jumper after cable pull test

Figure 3: Generic MPO assembly after cable flex test and LC jumper after cable pull test.

Top

 

PREVENT UNNECESSARY AND FUTURE EXPENSE

While where you buy optical fiber assemblies obviously impacts your bottom line, what you buy also matters. Within the data center environment, especially within switch-to-switch backbone links to the core and to the storage area network (SAN), there is immense pressure to handle extreme data volumes. To process this data, switch-to-switch optical fiber links are rapidly migrating from 10 Gb/s to 40 and 100 Gb/s and beyond. With several optical fiber types and applications available, data center managers need to understand which options reduce unnecessary initial upfront costs and prevent future expense.

First and foremost, data centers looking to upgrade their entire backbone data center cabling are faced with whether to deploy multimode or singlemode cable. While singlemode may offer the best future proof capabilities, the active equipment required remains more expensive than multimode equipment. Further, most data center backbone links do not require the reach distances currently supported by singlemode fiber, which include up to 10 kilometers (km) for speeds ranging from 40 to 400 Gb/s. On the other hand, large hyperscale data centers where backbone links often exceed the 100-meter (m) maximum link length supported by multimode equipment may want to stay abreast of developments in optimized-reach (i.e., 500 meter) singlemode data center solutions as a singlemode deployment could turn out to be their most cost-effective option.

Even selecting multimode fiber has become a more complex endeavor, especially with the introduction of the latest type of wideband multimode fiber (WBMMF), designated as OM5. While existing OM3 and OM4 multimode fiber is specified to operate in the 840 to 860 nanometer (nm) wavelength range with 850nm as the optimal wavelength, new OM5 specifies a wider range of wavelengths between 840 and 953nm to support wavelength division multiplexing (WDM) technology. WDM technology multiplexes multiple signals onto a single fiber using different wavelengths.

Fully backwards compatible with existing OM4 fiber specifications, OM5 fiber, and WDM technology offer the theoretical capacity to either increase transmission speeds or reduce fiber strand counts by a factor of four. For example, using standard OM3 or OM4 multimode fiber, 100 Gb/s require the use of 8 fibers via the 100GBASE-SR4 application (i.e., 4 fibers transmitting at 25 Gb/s and 4 receiving at 25 Gb/s). In contrast and as shown in Figure 4, using the single-lane 25 Gb/s application 25GBASE-SR with WDM technology, OM5 fiber links could potentially support 100 Gb/s using just 2 fibers (i.e., 4 different wavelengths on 1 fiber transmitting at 25 Gb/s on and 4 different wavelengths on 1 fiber receiving at 25 Gb/s on). Similarly, using 100GBASE-SR4 specifications, 400 gigabit OM5 fiber links could be created using 8 fibers transmitting and receiving at 100 Gb/s over 4 different wavelengths.

Figure 4: 25GBASE-SR with WDM technology over OM5 fiber links can support 100 Gb/s using just 2 fibers

Figure 4: 25GBASE-SR with WDM technology over OM5 fiber links can support 100 Gb/s using just 2 fibers.

While OM5 may seem like an advantage in terms of reducing fiber strand counts, it is important to note that there are no applications currently under development within the Institute of Electrical and Electronics Engineers (IEEE) to operate over this medium and, as a result, there is no available information on data rate, link length, or strand count for installing this media today. In addition, because IEEE typically develops applications based on a significant installed base, it is not certain when and if any application will be developed. Further, OM5 carries a significant cost premium over OM4, and a premium will also apply to any future transmission equipment. As such, anyone looking to future proof for 400 Gigabit might want to carefully weigh their fiber choice and consider one of the emerging singlemode fiber applications. For example, the pending IEEE P802.3bs (400GBASE-DR4) standard is slated to cost-effectively support 400 Gb/s over singlemode to 500 meters using 8 fibers with 4 fibers transmitting at 100 Gb/s and 4 receiving at 100 Gb/s. As previously mentioned, there is also work on singlemode technologies for shorter reach (500m) applications via the pending IEEE P802.3cd (100GBASE-DR) and IEEE P802.3bs (200GBASE-DR4/400GBASE-DR4) that may provide yet another case for singlemode fiber to be considered.

Speed

IEEE Application

Fiber Type

# of Fibers

Interface

Distance

10 Gb/s

10GBASE-SR

Multimode OM3/OM4

2 (1f x 10G)

Duplex LC

300m/400m

25 Gb/s

25GBASE-SR

Multimode OM3/OM4

2 (1f x 25G)

Duplex LC

70m/100m

40 Gb/s

40GBASE-SR4

Multimode OM3/OM4

8 (4f x 10G)

8/12F MPO (8)

100m/150m

50 Gb/s

50GBASE-SR1

Multimode OM3/OM4

2 (1f x 50G)

Duplex LC

70m/100m

100 Gb/s

100GBASE-SR21

Multimode OM3/OM4

4 (2f x 50G)

2 x Duplex LC

70m/100m

100 Gb/s

100GBASE-SR4

Multimode OM3/OM4

8 (4f x 25G)

8/12F MPO (8)

70m/100m

200 Gb/s

200GBASE-SR41

Multimode OM3/OM4

8 (4f x 50G)

8/12F MPO (8)

70m/100m

400 Gb/s

400GBASE-SR161,2

Multimode OM3/OM4

32 (16f x 25G)

32F MPO (32)

70m/100m

100 Gb/s

100GBASE-DR1

Singlemode OS1/OS2

2 (1f X 100G)

Duplex LC

500m

200 Gb/s

200GBASE-DR41

Singlemode OS1/OS2

8 (4f x 50G)

8/12F MPO (8)

500m

400 Gb/s

400GBASE-DR41

Singlemode OS1/OS2

8 (4f x 100G)

8/12F MPO (8)

500m

  1. Standard in development
  2. Likely too impractical to be commercially feasible/accepted

Table 1: Ethernet Optical Transceiver Roadmap includes multimode and singlemode fiber applications that are always divisible by either 2 or 8 fibers.

Those choosing optical fiber for the data center should also consider the pending IEEE P802.3cd (50GBASE-SR) standard — anticipated to release in 2018 — that will support single lane 50 Gb/s over OM3 or OM4 fiber. This standard demonstrates IEEE’s commitment to the development of higher bandwidth applications over the installed base of multimode fiber, and may make the case for avoiding the expense of a complete optical fiber plant upgrade to OM5 multimode or single mode.

Top

 

CONSIDER THE COST OF WASTED FIBER

In addition to the type of optical fiber, the MPO connector interface you choose can also impact your bottom line. In looking at current and future applications — for both multimode and singlemode — it is easy to see that the foreseeable future will be dominated by two and eight fiber solutions. Table 1 clearly shows that the Ethernet Optical Transceiver Roadmap includes fiber applications that are always divisible by either two or eight  fibers. However, most of the MPO connectors on the market today are 12-fiber connectors, and many data centers already have 12-fiber MPO backbone trunk cables in place that were originally deployed to support duplex optical fiber applications (i.e., 10GBASE-SR) using cassettes or modules that break out the 12-fiber trunks to six duplex connectors.

For applications like 40 Gb/s (40GBASE-SR4) and 100 Gb/s (100GBASE-SR4) that are based on eight multimode optical fibers, as well as future 400 Gb/s, the use of 12-fiber MPO solutions means that four fibers or 33% of the optical fiber goes unused. Not exactly cost effective. One way that data center managers can ensure 100% utilization of optical fiber with 12-fiber MPO/MTP solutions is to use conversion cords or modules that transition two 12-fiber or one 24-fiber trunk from backbone cabling to three 8-fiber MPO/MTPs for connecting to 40 and 100 Gb/s equipment. This is ideal for those data centers that already have 12-fiber or 24-fiber MPO-based backbone trunk cables in place, but it is not the most cost-effective option for anyone deploying new MPO-based plug and play fiber solutions that need to support eight-fiber applications, either now or in the future.

Not only does the use of conversion cords and modules add cost and complexity as a way to achieve 100% fiber utilization, but they also introduce some performance issues. When using conversion cords that transition two 12-fiber MPO to three eight-fiber MPO, three equipment ports would need to be taken offline in the event of a damaged cord — once again here one needs to consider the cost of downtime. The use of conversion modules also introduces additional insertion loss into the channel due to an extra mated pair, which as discussed previously is a concern in high-speed 40 and 100 Gb/s applications that have more stringent insertion loss requirements and could limit the use of cross connects. A much more efficient and cost-effective method would be to deploy eight-fiber MPO solutions (Base 8) that support current and future duplex and eight-fiber applications without wasting any fiber and eliminating the need for conversion cords or modules.

Top

 

REMEMBER COST IS MORE THAN A PRICE TAG

Carefully weighing the options and considering your future growth and application assurance for the future should also not be overlooked. If what you spend your IT budget on today adds no value tomorrow, or ends up costing you more because you are not able to effectively support your future needs, was it really the most cost-effective choice? Data center managers would be wise to not only examine their source vendors, but also rely on the expertise of designers and consultants who have a pulse on the available optical fiber options, applications and future standards developments, and can help you choose solutions today that will ultimately cost you less tomorrow. 

Editor’s note: This article originally appeared in ICT Today.

Top