Intelligent edge for sustainable agriculture Part 2: building sustainable networks with open source

In our previous blog we discussed the TMForum Intelligent edge for sustainable agriculture - phase II catalyst project, outlining how farmers can make better and informed decisions when given access to real-time data and insights. The blog also discussed how to leverage a unified cloud-native application platform from the core to edge, as shown in Figure 1, with a particular focus on integrating artificial intelligence/machine learning (AI/ML) solutions at the edge for farm operations to facilitate the move towards precision agriculture.

This blog focuses on sustainable computing, which actually applies uniformly across the entire network. While this is an example specific to farming to demonstrate what technology can achieve, the key findings can be applied to many other industry verticals.

Figure 1: A unified cloud-native application platform from the core to the edge

A key objective of this catalyst project was to positively impact society, and in particular sustainability requirements as outlined in the 17 United Nations (UN) sustainable development goals (SDGs). Our plan was to create a solution with a sustainability-first design principle.

One key aspect of the catalyst project was to embed pioneering solutions and implementations into the catalyst blueprint and infrastructure, especially at the edge, with a key focus on reducing energy consumption and carbon emissions.

In this blog we will outline how open source was used to measure and report energy efficiency and intensity, and for that data to be used to take meaningful action.

Sustainable farming

As part of phase II of the intelligent edge for sustainable agriculture catalyst project, edge AI and sustainable agriculture use cases focused on crop and animal health, and covered 9 of the 17 UN SDGs. As identified from our previous blog, an example of the sustainability goals achieved through these precision agriculture use cases was a 20%-30% reduction in water and pesticide usage.

Like every enterprise, farmers are under increasing pressure to produce sustainable products, including crops, while maintaining profitability and other corporate goals. Measuring and using affordable and clean energy sources, along with traceability of carbon footprint within the overall supply chain is becoming increasingly important.

The use of the latest technology and innovations allows farmers to plan for long-term sustainability, and meet their own sustainability goals as well as those mandated by governments and other bodies.

Measuring and reporting energy efficiency of the cloud-native network infrastructure

Sustainable computing is a multidimensional challenge for service providers and ecosystem partners. The Shift Project reports the information and communications technology sector global emissions from cloud computing accounts for 2.5% to 3.7% of all greenhouse gas emissions. According to TMForum, the telecommunications industry uses a staggering 2% to 3% of the total power consumption of humanity. Measuring, reporting and optimizing end-to-end implementations through sustainability-first design principles for cloud and network computing is a global imperative.

Energy consumption is increasing rapidly, especially as the industry accelerates to the edge. Any solution a service provider wants to deliver to an agriculture customer must have a positive sustainability impact. There is growing pressure to establish sustainability goals demanded by different stakeholders, from generations more conscious of their impact on the environment to governments aligning to protect the planet.

As part of phase I of the catalyst project, edge computing was identified as a rapidly emerging technology that requires transparent and open data to observe, measure, report, analyze and automate to achieve sustainability goals. This starts with driving awareness in and out of system architectures, as standards do not currently exist, and this is one area where open source communities excel.

Figure 2 shows a couple of open source projects that were applied to this catalyst project. Project Kepler provides power monitoring, and data aggregation and integration for observability. The Open Data Hub is used as an AI framework for analysis and reporting on sustainability.

Kepler was implemented within the catalyst project to measure energy efficiency and intensity in the cloud-native infrastructure (edge and multiaccess edge computing (MEC)), workload and application layers. Kepler uses eBPF to probe CPU performance counters and Linux kernel tracepoints. These data and statistics are fed into a machine learning (ML) model to estimate power consumption from the associated open source cloud-native application container orchestration platform Kubernetes pods.

Power consumption statistics are presented as Prometheus metrics and telemetry that can be used for pod scheduling or scaling, energy consumption reporting and visualization, and can be extended with carbon intensity metrics to report on carbon footprint of a cloud-native workload.

Figure 2: Power consumption monitoring with Kepler and integration with observability and AI reporting

As part of the catalyst project, we supplemented the use of Kepler by leveraging edge and device edge capabilities, and enhancing the models with generative AI (GenAI) to estimate power consumption to make it consumable for farmers. Red Hat OpenShift AI provides the analysis and reporting for sustainability, along with an open source-based workbench and AI/ML framework that can seamlessly integrate GenAI.

An example of this is a new model where GenAI embeddings are processed once and stored until they are used, allowing a graphics processing unit (GPU) to reduce its overall power consumption. Additionally, crop health-related data is combined with other data including third-party sources, integrated with GenAI and a ChatBot for the farmer to explore government sustainability requirements and incentives from related tax credits. We will dive deeper into how we applied AI/ML and in particular GenAI in several areas in our next blog.

Optimizing energy efficiency

As shown in Figure 3, we believe having a holistic approach is the right strategy, with fundamental capabilities of measuring common metrics, observability, reporting and analysis in place to focus on optimizations of power and energy efficiencies. Specifically for this catalyst project, our focus was on the edge, but the same approach and other sustainability focused solutions can be targeted at the 5G Core and radio access network (RAN) or anywhere in a cloud-native environment.

Figure 3: A holistic approach to achieving sustainable cloud-native networks

Energy optimization techniques within the service provider network, and enterprise and operational technologies can be enhanced with the aforementioned statistics, along with the identification of energy sources that include renewables and fossil fuels.

By augmenting statistics considered by the optimizers with information pertaining to CO₂ emissions, different techniques can be used to determine optimizations to apply within four network dimensions:

• Node level - These are optimizations that can be achieved through innovations of compute architecture. For example, power consumption by CPUs, smart network interface cards (SmartNICs), fine tuning of CPU core and memory frequencies, with the disabling unused cores, hardware accelerators and GPUs.
• Cluster level - These are optimizations that are available or applied holistically to a cluster. For example, energy-aware schedulers and de-schedulers, energy aware clusters, or pod auto scalers.
• System level - This represents the entire clusters and elements involved in delivering a service. For example, all the clusters, switches, routers and antennas that comprise the mobile network.
• Domain level - This refers to a specialized service or system functionality, for example, the radio access network (RAN) domain and the MEC domain. Each of these services has optimizations that are unique to the domain in which they operate. For example, in the case of RAN, the energy consumption of antennas can be reduced by adjusting the energy to individual sectors, or enabling and disabling them based on utilization characteristics. In the case of MEC, the optimization can focus on the location of the workload based on the type of energy source, or the overall utilization of a node.

Leveraging a holistic energy and carbon metric model for workloads from the device edge back to the core and cloud is a powerful approach to this problem.

Closing remarks

At TMForum Digital Transformation World this catalyst won the award for outstanding sustainability and impact on society, contributing to 9 of the 17 UN SDGs, with a unified cloud-native platform and GenAI playing key roles. The UN SDGs impacted include no poverty, zero hunger, clean water and sanitation, decent work and economic growth, responsible consumption and production and life on the land.

Specific to the solution work on Kepler in the sustainability category, the catalyst project addresses the following UN SDGs:

• Affordable and clean energy - leveraging project Kepler provides the capability for increasing power efficiency at the edge of the network.
• Industry, innovation and infrastructure - the cloud-native network has the capability to measure and scale for energy-efficient placements across the infrastructure, workload and application layers.
• Climate action - collection of CO₂ data from all parts of the value chain will allow farmers to make informed decisions, giving them the ability to reduce their carbon footprint up to 30% by 2030, with the catalyst giving them a blueprint to reach net-zero before 2050.

In addition to working with Kepler, Red Hat participates in a number of other Cloud Native Computing Foundation (CNCF) open source sustainable computing projects including the Kubernetes-based event-driven autoscaling (KEDA), container level energy-efficient VPA recommender for Kubernetes (CLEVER), power efficiency aware Kubernetes scheduler (PEAKS) and the Environmental Sustainability Technical Advisory Group. Red Hat also participates in many other related open source communities working on these UN SDGs, for example, Linux Foundation's OS-Climate and Enterprise Neurosystem and its Agriculture Innovation Mission (AIM) for Climate.

Through the power of open source projects that gather diverse contributions, and technology maturity that leads to global adoption, Red Hat aspires to positively impact sustainability goals and climate change. In this catalyst project we implemented projects and platforms to demonstrate how observability, measurement, reporting and analysis can be used to optimize energy consumption to reduce carbon footprint and greenhouse gasses.

With Red Hat OpenShift, an end-to-end unified cloud-native application platform, service providers can deploy an environment that has a sustainability-first design with embedded power monitoring. OpenShift AI has the capabilities to report, analyze and optimize solutions with a powerful AI/ML framework, and is an ideal open source workbench for introducing and expanding solutions with GenAI.

In our subsequent and final blog in this series we will discuss the role of GenAI, its benefit for enterprises to enhance sustainability and usability to capture intent for real-time feedback to help make more informed decisions.

To learn more see the Red Hat OpenShift product page and read the resources located at the TMForum website, including their blog, How to combat food insecurity with flexible deployment of 5G, AI and edge solutions in agriculture.