With an explosion of connected devices, there is no shortage of data being transmitted and stored via optical networking in the data center. However it’s not just the amount of data—it’s also how that data is being used. And that can be summed up with one word – sharing.
Ever-growing data sets are being shared across multiple vendor applications. In the massive high-density virtualized environments of cloud computing, this is driving more east-west server traffic. As the age of the “Internet of Things” come to fruition, the I/O portion of the equation will evolve like never before.
With all this change on the horizon, what does this mean for the cabling infrastructure inside the data center? Should it be all singlemode or is OM4 multimode the best bet?
Unfortunately there is no universal solution; every data center is unique. With newer discussions surrounding technologies like OM5 multimode fiber starting to gain ground, the balance between short- and long-term needs is becoming a complex situation. While we don’t have a crystal ball, here is what we can currently surmise:
This is where fiber strand cost is typically weighted heavier than termination and transceiver costs. Singlemode remains the clear choice in this category.
Fiber strand cost still weighs heavily in this group with termination cost and transceiver cost playing a larger role. For lower speed applications like 10G, OM4 is a good choice. However, looking forward to newer 25G VCSELs and even 40G-SR4, reach becomes a significant challenge. At 100G speeds, parallel singlemode using PSM4 technologies show significant promise. For both short and long term, singlemode is a solid consideration.
In this region, transceiver cost can still absorb some modest increase to decrease cabling and connectivity costs. 10G duplex architectures on OM3 have done well in this range. 25G duplex architectures on OM3 (and OM4 for the longer end of the range) are poised to become dominant, but adoption of OM3/OM4 for higher fiber count 40G-SR4 shows signs of struggling as the cabling infrastructure becomes a significant cost element.
The popularity of 40G BiDi implementations using duplex multimode leveraging WDM techniques is a clear indicator that some extra transceiver cost to minimize cabling infrastructure is a good fit in this range.
Looking ahead to 50G, 100G and 400G in this range, Ultra Wide Band (UWB) again extends WDM techniques to minimize fiber count, however the extra bandwidth needed introduces the OM5 multimode discussion. Here is where we need to move with caution, carefully balancing short- and long-term needs—OM3 (<70m) and OM4 (70-100m) is the short-term solution while long-term could mean OM4 for shorter lengths and OM5 for longer lengths becoming dominant to avoid the exploding fiber count that ultimate caused the 100G-SR10 (MPO-24) application’s failure to launch.
Here transceiver cost outweighs fiber and connector cost, so minimizing the cost (and power) of the transceiver are paramount. Active Optical Cable (AOC) assemblies that include embedded optics and eliminate connectorization are dominant here in the short term where fiber grade, type and count become irrelevant.
Alternatively, cost-effective transceivers today are VCSEL based which ultimately drives OM3 multimode fiber. Longer-term optical integration leveraging silicon photonics is the game changer that will introduce new connectors and potentially spawn a new generation of cost-effective, shorter-reach singlemode and multimode options.
As you can tell, the future of fiber is rather complex and is a rapidly changing landscape. Comparing the future outlook from just a year ago with the future outlook today is rather telling.
For example, when we consider the changes that have happened in the past 6 months, we can easily say that if we incurred the additional expense last year to “future proof” with MPO-24 connectors, we would now be in a bad position for tomorrow. Conversely, let us not underestimate the ability of our industry to adapt to the brownfield infrastructure.
Dwayne Crawford has more than 20 years of experience in the datacomm industry. He has served on several international standards committees to advance high-performance/low-latency protocols (such as IEEE-1394, GigE Vision and CameraLink) used in real-time image processing and utilizing high-performance computing platforms.