logo_header
  • Topics
  • Research & Analysis
  • Features & Opinion
  • Webinars & Podcasts
  • Videos
  • Dtw

Cosmote trials AI-driven NFV energy savings to deliver more sustainable 5G edge computing

As part of a TM Forum Catalyst, Greek operator Cosmote tested Intracom Telecom's NFV Resource Intelligence platform (NFV-RI), which is designed to significantly reduce the energy consumption of an NFV datacenter by dynamically and automatically adapting the power of dataplane Virtual Network Functions (VNFs) to the actual load.

Joanne TaaffeJoanne Taaffe
16 Nov 2021
Cosmote trials AI-driven NFV energy savings to deliver more sustainable 5G edge computing

Cosmote trials AI-driven NFV energy savings to deliver more sustainable 5G edge computing

Who: Cosmote and Intracom Telecom

What: An AI-driven platform that saves energy by calculating and applying efficient ways to manage power consumption within NFV datacenters without impacting service levels.

How: As part of a TM Forum Catalyst, Greek operator Cosmote tested Intracom Telecom's NFV Resource Intelligence platform (NFV-RI), which is designed to significantly reduce the energy consumption of an NFV datacenter by dynamically and automatically adapting the power of dataplane Virtual Network Functions (VNFs) to the actual load.

Results: Reduction in servers’ energy consumption by 14% on average over a 24-hour period with a peak saving of 35% during the lowest periods of traffic, without impacting customer service.

In 2018, data centers alone accounted for 2.7% of the electricity demand in the EU28, according to a recent EU report. The same report expects the energy consumption of data centres to increase by 21% by 2025 as the demand for digital services grows. As communication service providers (CSPs) around the world deploy more 5G services, they will need energy efficient technologies to help them offset the increases in power consumption.

The roll out of 5G networks not only promises to increase demand for new digital services. It is also helping drive uptake of network functions virtualization (NFV) technology. International Data Corporation (IDC) forecasts worldwide revenue for telecommunications network functions virtualization software, including virtual network functions (VNFs), network functions virtualization infrastructure (NFVI), and cloud-native network functions (CNFs), will grow from $7.5 billion in 2020 to just over $29 billion in 2025.

NFV helps CSPs reduce total cost of ownership and increase their service scalability and agility, including when they expand mobile networks from the network core to the edge. However, despite its benefits, NFV presents significant operational efficiency challenges. For example, dataplane VNFs, such as user plane function (UPF) in the 5G core, rely on aggressive polling to achieve their stringent KPIs, which include deterministic performance, low latency and zero data packet drops.

Managing power consumption on the edge

As a result, the CPUs in VNF servers are kept running at the highest possible frequency. This means that even where there is zero or very low traffic on the network, the servers consume the maximum possible amount of power. As CSPs accelerate their 5G deployments, they can expect to deploy edge servers within each typical metropolitan area. Even a potential 5% reduction in the power consumption of edge servers would directly translate to significant OpEx reduction. It would also help CSPs comply with growing consumer and regulatory demand for sustainable communication services.

Greek operator Cosmote, part of the Deutsche Telekom Group, worked with Intracom Telecom to look at how it could lower the power consumption of VNFs by making it possible for their CPUs to automatically switch to running at lower frequencies during dips in network demand, while still meeting its service level objectives.

Cosmote and Intracom Telecom tested Intracom’s NFV Resource Intelligence platform (NFV-RI) on a prototype of its mobile network as part of a trial within the TMF Catalyst, "AIOps Autonomous Service Assurance". NFV-RI uses AI to automatically reduce the energy consumption of the NFV datacenter by dynamically adapting the power of dataplane VNFs to the actual traffic load.

The results convinced Cosmote that AI-based management solutions will play an important role in creating more sustainable communication services.

“Achieving energy savings by adjusting server power in line with traffic, fully automatically and in real-time, showed that AI-based management solutions suggest the only way towards sustainable communication services, according toDimitrios Simantirakis, CEM Network Engineer, Cosmote.

Demystifying black boxes

The platform works by employing AI in a closed loop to predict what load VNFs will have to cope with at any given moment. The system then automatically adjusts power consumption to match data traffic on the server, while still respecting service level agreements. The system also uses AI and rich telemetry to gain visibility into closed, "black-box" VNFs to accurately assess what is the lowest amount of power they need to sustain their traffic without dropping packets.

The ability to work with black box VNFs was an important requirement. Ideally, the closed-loop mechanism would rely on VNFs to provide a real-time indication of how busy they are in processing packets, at any moment. However, many of the VNFs deployed on CSP networks are closed enterprise systems that are not set up to share key information, such as reporting of the load on their receive queues. Instead, they operate as de-facto black boxes that expose little of the information needed to make quick-fire, accurate decisions.

The NFV-RI therefore leverages low-level and generic platform metrics to indirectly infer how busy VNFs are at any moment. Using AI techniques and a machine learning procedure, it is able to calculate the minimum power level a specific VNF would need to sustain a certain traffic level without errors. This is a critical function because if there is not enough power to process data, then the CSP will lose packets and fail to meet its service level objectives.

Preventing an overload

The NFV-RI platform’s closed-loop mechanism involves deploying an AI software agent locally on every NFV server, which is able to accurately predict if a short-term traffic overload is about to occur.

It automatically calculates exactly when and how to increase CPU frequencies without wasting energy or dropping packets. Equally, the system is set up to detect underload situations so that it can reduce the power available to VNFs that are dealing with low traffic levels.

Cosmote and Intracom based the trial on Cosmote’s real-time mobile network traffic patterns gathered over a 24-hour period. They were fed into a virtualized Evolved Packet Core (EPC) prototype based on OMEC.

The trial focused on polling-intensive operations, namely the service and packet gateway functions (SGW-U, PGW-U). Using the NFV-RI platform, Cosmote was able to automatically align its server’s power consumption with the highs and lows of data traffic for the entire 24-hour period, thereby reducing power consumption to 227 Watts, which represented an overall saving of 14 %. The maximum power saving rose to 35% (196 Watts), during the night, when traffic density is lowest. The completely automated AI-based system is also able to self-heal if its predictions do not work and protects data traffic integrity by switching back to the default mode of operating the VNFs at maximum power.

The trial tested more than the NFV-RI platform’s ability to address algorithmic challenges. It was important that Cosmote could easily integrate the AI components into its operational environment and manage them as they would any other software component.

Intracom Telecom therefore used TM Forum’s AIOps Service Management Framework (IG1190) to structure the technical and operational processes needed to deploy and integrate AI components and their relevant business capabilities into the CSP’s existing IT and network operations.

For example, TM Forum’s AIOps Monitoring and Event Management (IG1190F) process makes it simpler for production teams to prevent operational issues and provides mechanisms for the early detection and resolution of incidents, as well as to filter and correlate events. AIOps Incident Management (IG1190G), meanwhile, focuses on restoring the normal services as quickly as possible and in line with the agreed SLAs, thereby minimizing the adverse impacts on business operations.

The NFV-RI has also started using advanced forms of AI like reinforcement learning to create cognitive agents that are able to learn by themselves to make optimal resources decisions in a live telco network and share the knowledge between them across the multiple edge locations.

Reducing consumption without impacting service

Using NFV-RI, Cosmote was able to adjust the power consumption of servers running virtual network functions (NVF) in line with traffic. Normally the servers’ CPUs run constantly at the highest frequency to guarantee optimal service delivery, regardless of actual demand. As a result of using NFV-RI, Cosmote reduced the servers’ energy consumption by 14% on average over a 24-hour period with a peak saving of 35% during the lowest periods of traffic, without impacting customer service.