Notes from Nice: 400G+, Disaggregation, 5G and AI/ML
I recently had the good fortunate to attend the NGON & DCI Europe event that was held in Nice, France on the 26th-28th June. This event always provides an excellent opportunity to take the pulse of the optical networking industry with a good mix of network operators, system vendors and component suppliers, plus the occasional academic and one or two ICPs. Looking back over three days of presentations, panel sessions and meetings, a number of key themes emerged:
A number of presentations looked at both high speed line interfaces and high speed clients. Line interface topics included higher baud rates and novel modulation techniques, including constellation shaping, enabling wavelengths of between 400G and 600G depending on the vendor. The OIF talked about their efforts to standardize technologies for 400G+, following the success their standardization efforts have had in speeding the adoption of coherent 100G. These current efforts include specifications for high baud rate (45GBaud, 64GBaud) analogue pluggables and 400G ZR which aims to create a compact and cost-effective option for transmitting 400G over DWDM at distances of up to 80km. On the client side topics included the form factor for 400G client pluggables with QSFP-DD seen as the favourite due to its smaller size and lower power but with OSFP, less challenging for manufacturers, also in the race. Flex Ethernet (FlexE) which provides more flexibility in matching line capacity and client capacity while avoiding the ~20% inefficiency of LAG (Link Aggregation) was also discussed, with the 2.0 version adding the ability to use Nx200G or Nx400G in addition to Nx100G of the 1.0 version, while also providing a 25Gbps granularity option in addition to 5Gbps granularity of the 1.0 version.
While a couple of years ago disaggregation and more specifically open line systems were novel and somewhat controversial concepts, at least outside the DCI stream of the conference, and primarily promoted by system vendors, this year disaggregation went mainstream. From the research and education community, GÉANT, the pan-European backbone, and CSC/Funet in Finland discussed their experiences of deploying disaggregated muxponders over their existing DWDM line systems. LINX, the London Internet Exchange, talked about their experiences with both disaggregated muxponders and disaggregated switch/routers based on white boxes and a hardware-independent NOS, which was also the topic of a presentation from my Coriant colleague, Harald Bock.
Slide on Disaggregation presented by Coriant’s Harald Bock
Telefonica coined the term “partial disaggregation” to describe their preferred approach of keeping the all the components of the optical line system with one vendor and the muxponders with different vendors, as opposed to “full disaggregation” that would mix line system components such as amplifiers and WSSs from different vendors. And Orange talked about the TransportPCE project which is working to develop an open, multi-vendor SDN controller for Open ROADM-based infrastructure.
Triangle of 5G Applications (Source: ITU-R IMT 2020 Requirements)
5G was yet again a hot topic. Almost every presentation on 5G included a slide with a triangular graphic, based on the one from the ITU-R IMT 2020 Requirements shown above, depicting the three key objectives of 5G:
- Enhanced Mobile: up to 10Gbps to user devices, with 1000x bandwidth per unit area
- Massive IoT Scale: huge increase in IoT devices with ultra-low power enabling 10 year batter life
- Ultra-Reliable and Low Latency: for applications such as autonomous driving, remote surgery, etc.
The consensus view was that building a single network that could simultaneously meet all three objectives would be technically very challenging while building three networks would not be economically feasible. There was a lot of optimism that optical had a key role to play in terms of delivering the capacity and low latency requirements, with one presenter excitedly pointing out that 5G base stations today have between one and four 10G ports, with multiple base stations per tower. This will require a lot more fiber in the fixed access network but it also creates requirements for more spatially-efficient fiber, the topic of a presentation from Corning, and more cost-effective optical equipment.
BT talked about the EU Metro-Haul project it is leading, to architect and design energy-efficient programmable metro networks for 5G, a project in which Coriant is also a participant. One topic of debate was best approach for cost-effective, high speed transmission in the backhaul/fronthaul networks: cost-optimized, temperature-hardened coherent technology (i.e. coherent-lite) or direct detect approaches such as PAM4 or DMT. Finally, Heavy Reading shared the results of a recently conducted survey that showed latency as being the key 5G transport network challenge for mobile operators, a slight preference for traditional backhaul and a distributed RAN (DRAN) vs. fronthaul and a centralized RAN (CRAN), and strong roles for both packet over fiber and point-to-point WDM.
Artificial Intelligence/Machine Learning
A final theme looked at the potential roles of artificial intelligence (AI) and machine learning (ML) in helping to automate optical networks. Intent-based networking was covered in a number of presentations: you tell the network what you want (i.e. 50Gbps between A and Z with 99.99% or better availability and 10ms or lower latency) and it figures out how to get it done. My colleague Stefan Voll’s presentation included a number of practical examples of how AI/ML could be applied to optical networks: advanced warnings of fiber breaks based on vibration detection, predicting equipment failures before they happen, predicting traffic congestion, enhanced path selection, and network defragmentation.
Slide on AI Applications for Optical Networks presented by Coriant’s Stefan Voll
Another example he gave was dispatching engineers in advance to areas likely to be impacted by storms. LINX saw alarm correlation as an area where AI/ML could add significant value. The growth in optical network scale and complexity driven by 5G was seen as a key driver for AI/ML, though the Heavy Reading survey showed automation as being currently low on the list of 5G transport network priorities. BT also cautioned that traditional linear learning might still be sufficient for may use cases.