It would be great if we could flip a switch on the networks that currently consume about 3% of the world’s power and emit roughly 2% of the world’s greenhouse gasses to make them more sustainable. And I’m sure any telecommunications service provider would want to reduce their network operating costs by being more energy efficient. But achieving sustainability goals is a complex and layered problem, and one that requires an entire ecosystem to work together to reduce the telecommunication industry’s carbon dioxide footprint. 

Many service providers are looking to advanced data analytics powered by artificial intelligence (AI) to drive better outcomes. However, AI can only be as good as the data we feed it, and collecting enough accurate data in areas like traffic patterns and energy consumption across both traditional, cloud and multi-vendor networks is a challenge still to be solved. There is also the challenge of converting AI recommendations into real time manipulation of the network and associated workflows, by means of massive automation. On top of that, AI itself can be a significant contributor to power consumption so we need to be very deliberate in its use for network power optimization. 

Even today, many sustainability projects are implemented in isolation, focusing on individual network domains or challenges and this approach has shown its limitations. In order to maximize the impact of energy efficient strategies, a more holistic approach is required across network domains by leveraging open platforms that use advanced data analytics, AI and automation. I want to use this post to outline some of the building blocks of a unified approach, highlighting some combined efforts Red Hat is actively working on with our partners.

Open 5G core infrastructure for maximized value and reduced power consumption

Red Hat, NEC and Intel have partnered together to deliver an open 5G core infrastructure that helps reduce the operational expenditure associated with power cost and consumption. Now, NEC can reduce commercial power consumption of NEC converged 5G user plane function (UPF) on Red Hat OpenShift by more than 30% with Intel® Infrastructure Power Manager for 5G Core. With Intel® Infrastructure Power Manager for 5G Core, every processor core has the right power at the right time to keep emissions and costs to a minimum. Learn more about how Red Hat, NEC and Intel are working together to contribute to sustainability and power reduction for 5G networks.

Sustainable call processing and packet continuity with greater power

Last year, we showcased a collaboration between Red Hat and Arm to deliver more energy efficient 5G and vRAN solutions fueled by Red Hat open source technologies and Arm® compute platforms.

More recently, in collaboration with NEC, Arm and Qualcomm Technologies, Inc., we have successfully demonstrated end-to-end operation of NEC’s open virtualized radio access network (vRAN) and 5G core products using Qualcomm® X100 5G RAN Accelerator Cards and Arm Neoverse™-based CPUs on Red Hat OpenShift in a commercially equivalent environment. By integrating our technologies, we’ve been able to successfully demonstrate sustainable call processing and packet continuity with enhanced power and space savings with the potential to significantly reduce the total cost of ownership for service provider RAN deployments.

Using Kepler and AI to monitor power usage

Model training and inference consume substantial amounts of energy at the level of containers, pods, namespaces. The open source project Kepler, or Kubernetes-based Efficient Power Level Exporter, captures power consumption metrics across a wide range of platforms to help system administrators and developers understand, optimize and plan power usage. The technology, which was co-founded by Red Hat and IBM Research and is used by Red Hat OpenShift for power monitoring, can capture energy consumption by both CPU and graphical processing unit (GPU), thus providing insight into compute utilization patterns of training and inference tasks and evidence for further tuning and optimization.

Kepler helps equip other open source projects and technologies with the data they need to better manage energy. An example is SusQL, an open source project used to track AI model training jobs in distributed environments, which uses Kepler metrics to aggregate power consumption of distributed training jobs, extending the power consumption insight into the cluster level. 

It also enables optimization model inference services, like those on Red Hat OpenShift AI, an MLOps platform for building, training, deploying, and monitoring AI-enabled applications on OpenShift., Together, power consumption can be monitored to derive energy usage patterns. These patterns can be correlated with system configuration and service provisioning to recommend optimal performance per watt configuration. With tools like Kepler integrated into Red Hat’s AI portfolio, we can help AI become more sustainable.

Energy efficient AI analysis at the edge

Red Hat is dedicated to our work within the IOWN Global Forum to help deliver smarter solutions for the future with sustainability in mind. Most recently, Red Hat, NTT, Fujitsu and NVIDIA were able to demonstrate an energy-efficient IOWN-based platform solution with NTT’s accelerated data pipeline for AI analysis services. Through optimizing AI inferencing at the edge for large scale video camera data analysis, we were able to realize a significant reduction in power usage by combining IOWN all-photonics network (APN) and data-centric infrastructure (DCI) with Red Hat OpenShift to deliver large-scale AI data analysis.

This collaboration demonstrates that it is possible to significantly reduce power consumption while maintaining low latency in the use case of video AI analysis at the edge. We found that even when a large number of cameras were connected, the latency required to aggregate and analyze the data with AI can be reduced by 60% as compared to latency with centralized clouds. By implementing container technology with Red Hat OpenShift, NTT is able to flexibly operate AI analysis processing with ease. This also proves that this solution can be widely applicable from video AI analysis in smart cities to similar scenarios with many distributed sensors.

Red Hat, Intel and Ericsson reduce network power consumption

Red Hat is collaborating with Intel and Ericsson to develop, integrate and deploy more sustainable cloud-native technologies that reduce energy costs and carbon emission of networks. There is a balance to strike in optimizing and reducing energy consumption while maintaining network performance. Red Hat, Intel and Ericsson are focusing initially on radio access networks (RANs) and have been able to demonstrate up to 20% savings in processing power consumption using 4th Gen Intel® Xeon® Scalable processors with vRAN boost. The solution has three pillars: hardware, including energy-efficient servers and accelerators; software, such as real-time dynamic scaling of CPU cores for the RAN distributed unit (DU) and centralized unit (CU) workloads; and automation: real-time, energy-aware automation that’s based on network utilization and traffic patterns. Watch this video to learn how Red Hat, Intel, and Ericsson are helping service providers build more sustainable cloud solutions.

Smart scaling for telco datacenters

Red Hat and Intracom Telecom are working to deliver a solution that optimizes OpenShift-based infrastructures (both on-prem and cloud-based), by proactively adjusting server workloads and strategically powering off underutilized servers to address idle power consumption, without compromising workload performance and stability. The number of worker nodes is dynamically scaled, based on real-time predictions of resource demand to ensure optimal infrastructure scalability and efficiency. This solution is well suited for converged telco data centers hosting 5G network functions (e.g. user plane function pods, control-plane pods) and associated services (AI pods, edge pods), as traffic load for these services fluctuates cyclically, following daily and weekly patterns. The smart scaling capability has demonstrated a potential to significantly reduce the average number of worker nodes, even in telco datacenters exhibiting mild load fluctuations within a day, thus delivering important energy and cost savings.

Collaboration can drive sustainable innovation

Red Hat brings to the table its experience in developing open source tools, capabilities and methodologies that make sustainability an integral part of the control and management of cloud-native architectures. We also bring our history of more consistent and open collaboration to galvanize a community of original equipment manufacturers (OEMs), independent software vendors (ISVs), customers and policymakers to promote industry-wide sustainability best practices. Data collection and analytics will be key for our customers to make the best decisions based on measuring the right things. Red Hat continues to look at enhancing observability capabilities in conjunction with AI platforms and massive automation to analyze data at scale and make in-time recommendations on actions to take. Together with our customers and partners, we can enable a holistic, data-driven, energy-saving approach across IT, network, edge, core, and cloud environments.