Skip to main content

What does it take to unleash the power of data in optical networks?

What does it take to unleash the power of data in optical networks?

Today’s wearable devices help us keep track of our health by collecting information on our heart rate, ECG, blood pressure, sleep patterns and more. This information is typically uploaded to the cloud, where systems perform further analysis and monitoring, or send us notifications when our health could be at risk. The keys to getting the most from wearables are to capture quality data and have tools that make it easy to monitor our health, diagnose potential illnesses and provide relevant information to healthcare professionals. High-capacity connectivity like that promised by 5G is also important for ensuring that data can be processed in a timely way.

These keys to success are essentially the same as those for solutions that aim to use data to monitor the health of an optical network and automate it to maximize performance and reduce operating costs. However, most optical monitoring networks, commonly referred to as Data Control Networks (DCNs), are built around bandwidth-constrained operations, administration and maintenance (OAM) channels that don’t currently offer a path to greater capacity.

The first DCNs were designed to support network management operations over low-speed, low-overhead, in-band management channels. They were not intended to collect large volumes of performance data to support applications that use new technologies such as machine learning to improve optical network performance.

Using YANG and gNMI to unleash the power of data

To unleash the power of data, Nokia has enhanced its optical network equipment with innovative streaming telemetry options that can reduce the bandwidth demands on the DCN. We have captured streaming telemetry data through a standardized, optical network-specific YANG model.

YANG is a data modeling language that can be used to define datastores. It was developed by the Internet Engineering Task Force (IETF) and published in RFC 6020. The language is protocol independent and can be converted into any encoding format that the network configuration protocol supports. YANG is a modular language that represents data structures in tree format and includes several built-in data types. Its data models can be extended with new data models or augmentations.

The data associated with the YANG model is then streamed using protocols such as Network Management Interface (gNMI) over Remote Procedure Call (gRPC). gRPC is an open source remote procedure call mechanism that simplifies the creation of distributed applications and services by enabling client applications to directly access server applications on different machines. It can be used to define services and specify the operations (methods) that can be called remotely with their parameters and return types. The server implements gNMI and runs a gRPC server to handle client calls. The client has a so-called stub that provides the same methods as the server.

gNMI is a unified streaming telemetry and configuration management protocol. It defines an interface that enables network management systems to interact with network elements. As the name implies, it is built on the open source gRPC framework and provides a single service for state management (streaming telemetry and configuration).

Today, gNMI and the YANG models are becoming the gold standard for streaming telemetry.

Customizing streaming telemetry for DCN performance

Nokia WaveFabric optical network equipment introduces innovative streaming telemetry support that reduces the bandwidth required to support data capture. These new capabilities make streaming telemetry work over lower-speed DCNs without impacting existing network management functions. They are described below with the aid of the following figure.

Figure 1

Flexible configuration of the reporting interval

To reduce streaming telemetry bandwidth requirements, WaveFabric makes a clear delineation between sampling and reporting data. The data sampling rate can be extremely fast (<50 ms), and the resulting data can be stored locally on optical network equipment. However, figures such as the data average, minimum and maximum can have a longer reporting period (e.g. one second or more). This preserves data quality while reducing the bandwidth impact on the DCN.

Furthermore, not all data requires the same reporting period for the purpose of analytics. For example, link transmission (Tx) power can tolerate a reporting period of 15 minutes, whereas a State of Polarization (SOP) requires a much shorter period to be meaningful. Nokia has carefully and individually designed sampling and reporting intervals for every key performance indicator (KPI) to maximize insight while minimizing impact on the DCN.

Suppression of redundant data including threshold management

Some data insights can be easily derived from other data. This means a significant amount of second-order information doesn’t need to be streamed from the network equipment. To achieve a given interest level, operators can establish a threshold range for which data are not streamed. These can be absolute or delta thresholds.

Both capabilities further reduce the data capture impact on the DCN. For example, in normal conditions, Tx power does not fluctuate very much, so data will be sent only when a threshold has been crossed. In addition, Tx power deviation can be calculated instead of being sent from the network element.

Sophisticated content filtering

Operators can also reduce data query overhead by adding XPATH filtering, a proposed standard for a filter-specification language for XML, when they access the YANG data model (e.g. NETCONF subtree filtering for gNMI). This also enhances the usability of information from the YANG model.  

Securely storing large amounts of data

Once network operators have efficiently captured data over the DCN, where do they store this data? Data retention mechanisms such as the traditional databases are becoming impractical for the long duration of retention required to make data correlation efficient and support machine learning. Operators must also have the flexibility to take advantage of cloud storage options to reduce the cost associated with storing and analyzing data.

Extracting insights from data

With reliable, efficient and secure data collection and storage mechanisms in place, together with easy-to-use tools such as those available from Nokia WaveSuite Network Insight applications, data scientists can analyze trends and use machine learning to automate optical networks.  

Join us for the next post in this series, in which Dr. Sébastien Bigo from Nokia Bell Labs sets the stage for how network operators can use what they learn from optical networks to enable automation that maximizes network performance and reduces network total cost of ownership (TCO).


Learn more about data analysis and machine learning

Sylvain Chenard

About Sylvain Chenard

IEEE senior member Sylvain Chenard received his BScA from the Département du Génie Electrique at Université Laval, Québec, Canada in 1994. He also received a technical degree in microelectronics from Collège de Sherbrooke, Québec, Canada in 1988. Sylvain is a member of the IEEE Communications and Computer Societies and the Ordre des ingénieurs du Québec. He is currently a senior product manager at Nokia with a focus on optical network analytics. He works closely with Nokia Bell Labs to define and productize state-of-the-art techniques and algorithms for estimating and monitoring transmission quality.

Article tags