This text is a follow-on to “5G Vitality Effectivity Metrics, Fashions and Methods Checks” which was printed by RCR Wi-fi Information on August 28, 2024
Introduction
The Aether SMaRT-5G initiative (initially began by the Open Networking Basis (ONF)), is targeted on analysis and growth of revolutionary vitality measurements and modeling for 5G cellular networks. The undertaking additionally contains demonstration and validation of superior vitality financial savings strategies in collaboration with a multi-vendor/operator ecosystem.
The Nationwide Telecommunications and Data Administration (NTIA) of the US Division of Commerce has been supporting the SMaRT-5G initiative with a analysis grant via their Public Wi-fi Provide Chain Innovation Fund (PWSCIF). The $2M grant is funding SMaRT-5G in collaboration with Rutgers College’s WINLAB to analysis, develop, and validate correct and efficient check strategies and to create metrics and fashions to measure the vitality effectivity of 5G community elements in addition to the effectiveness of end-to-end Open RAN based mostly vitality optimization methods.
The platform we’ve got created for this goal is named POET: Platform for O-RAN Vitality Effectivity Testing. Particulars of POET could be present in a beforehand printed RCR Wi-fi article.
We have now adopted completely different approaches for energy measurement, which is the primary and essentially the most basic step within the undertaking. Simply to recap, we use a multi-pronged strategy for this goal.
(a) PDU: The ability, present, voltage provided to PNFs and servers is obtained by recurrently querying the Energy Distribution Models (PDUs) (nominally each 15 secs), and exporting this knowledge to Prometheus/Grafana. This gives a vital ground-truth measurement of energy consumption.
(b) IPMI: Energy and surroundings variables reported by the servers are additionally monitored utilizing the Clever Platform Administration Interface (IPMI). Queries are made to the server Baseboard Administration Controller (BMC) and exported to Prometheus/Grafana.
(c) Scaphandre: We deployed the Scaphandre open-source vitality monitoring performance on bare-metal servers working CU/DU software program. Scaphandre measures course of utilization (based mostly on Intel Operating Common Energy Restrict (RAPL)) and estimates the ability consumption utilizing an estimation mannequin.
(d) Kepler: The Kubernetes deployment within the testbed makes use of Kepler exporters on every server sending energy metrics to Prometheus/Grafana. Kepler is an open-source vitality monitoring performance for Kubernetes programs. It measures node and container utilization (based mostly on Intel RAPL) and estimates the ability consumption utilizing an estimation mannequin.
Although PDU offers the bottom fact by way of vitality consumption, deploying PDUs in a community and gathering knowledge is pricey and laborious. Moreover, whole vitality consumption alone is inadequate to attain the specified targets of vitality administration efforts, therefore granular and detailed vitality consumption measurements are required. As an illustration, when a disaggregated RAN (such because the O-RAN structure) is taken into account, you will need to understand how vitality is consumed as a operate of visitors load by RU, CU, DU and Core, together with infrastructure overhead which homes digital community features. Due to this fact, if O-RAN interfaces can present vitality consumption data with enough particulars and granularity, from which varied vitality effectivity KPIs could be derived, that will be ideally suited.
Nevertheless, the present scope of O-RAN interfaces is insufficient for this goal, and we’ve got to depend on different possible and economical approaches to complement O-RAN interfaces, to start with. Bearing this in thoughts, different approaches like IPMI, Scaphandre, and Kepler are extremely related, and correlating knowledge collected by way of these strategies is essential. Efforts are actually underway to gather and analyze knowledge from POET alongside these strains. Preliminary outcomes had been not too long ago printed as a technical paper and introduced on the RitiRAN convention by undertaking contributors from the Linux Basis’s Aether SMaRT-5G undertaking in collaboration with WINLAB and Cognizant. This text summarizes the important thing ends in the paper.
5G Community Vitality Consumption – Key Observations
We have now noticed that PDU and IPMI readings observe nicely, and PDU readings sometimes differ round 5% above IPMI readings, relying on the load. Thus, IPMI readings could be calibrated to estimate PDU readings. That is precious within the context of vitality consumption of cloudified and virtualized community features. It’s also noteworthy that IPMI readings are quantized the place the readings are reported in steps, whereas PDU readings are steady. This must also be thought-about within the calibration/modelling efforts.
We used Kepler to get vitality consumption of Kubernetes containers/pods, as an try and measure the vitality consumed by Cloudified Community Capabilities (CNFs). Kepler makes use of fashions based mostly on CPU utilization to estimate vitality consumption, and the fashions can differ for various CPU/server households which could be tweaked to enhance estimation accuracy. The present outcomes are by utilizing open-source Kepler fashions (as-is) and we plan to check strategies to enhance accuracy of fashions. Understandably, Kepler readings are considerably beneath PDU/IPMI readings since they don’t seize non-CPU-related vitality consumption, which must be modelled and calibrated. Nonetheless, we noticed that Kepler readings observe the PDU and IPMI readings fairly nicely, with respect to whole vitality consumption.
You will need to see how the vitality consumption is for varied digital community features, specifically Core, CU and DU, as reported by Kepler. We did some TCP checks, first with one UE after which with two UEs. We made the next observations:
1. The ability consumption of the 5G Core, O-CU, and O-DU in a Kubernetes deployment could be individually estimated utilizing Kepler. The overall consumption of the Core NF, and CU and DU containers are solely part of the vitality consumption of the entire O-RAN system on account of important vitality consumption from different Kubernetes system features.
2. The impact of load with respect to vitality consumption is essentially the most on DU versus Core and CU.
We additionally seemed on the vitality consumption reported by Scaphandre (naked steel), and in contrast it with Kepler, PDU, and IPMI readings. Particularly, we used Scaphandre to get vitality consumption of O-RAN NF (OAI) and different processes. Additionally, like Kepler, Scaphandre additionally makes use of fashions to estimate vitality consumption based mostly on CPU utilization, and we plan to check strategies to enhance the accuracy of fashions.
Our key observations are summarized beneath:
1. Scaphandre readings tracks nicely with respect to Kepler, PDU and IPMI readings
2. The ability consumption of the OAI O-RAN system may very well be individually estimated utilizing Scaphandre. The overall consumption of the OAI O-RAN system is considerably beneath PDU/IPMI readings because the Scaphandre estimates don’t seize non-CPU associated vitality consumption, which must be modelled and calibrated
Efficiency Measurements
One vital goal for this R&D effort is to correlate vitality consumption to community efficiency and to derive KPIs, connecting each these points. Due to this fact, efficiency measurements are additionally vital on this context. We have now been adopting varied approaches for this goal, comparable to:
1. Finish-to-end measurements (for instance, knowledge volumes and throughput)
2. O1-based KPIs (we used the telnet-based OAI DU O1 resolution to get uplink/downlink throughput, and downlink PRB load and anticipate extra assist for O1 in future checks).
3. E2-based KPIs (A number of KPIs out there over E2 interface, e.g. E2-SM KPM)
Subsequent Steps
POET is operational now, and preliminary observations point out that IPMI is a helpful estimate for whole server vitality consumption, whereas Kepler, Scaphandre are promising instruments for O-Cloud VNF/CNF energy estimates. At the moment, the workforce is within the strategy of calibrating varied outcomes, augmenting RU energy measurements utilizing business RUs, and correlating energy consumption measurements with efficiency measurements to develop energy consumption KPIs and fashions. As well as, dialogue with different labs is underway, particularly ones funded by NTIA (comparable to ORCID and ACCoRD), with the purpose of increasing experiments and getting nearer to real-life community deployment eventualities.
Register to attend the digital Telco Sustainability Discussion board, hosted by RCR Wi-fi on December tenth. Sarat Puthenpura will likely be taking part in a energetic panel dialogue “From the core to the RAN – the way to make networks extra sustainable”.
In regards to the creator
Sarat Puthenpura, Ph.D., is the Chief Architect – Open RAN, SMaRT-5G and Aether tasks hosted by the Linux Basis. He has over 35 years of expertise within the business spanning the applying of mathematical optimization and synthetic intelligence/machine studying strategies in telecommunications community planning, useful resource optimization, operations automation, and efficiency administration with management in R&D, software program growth and deployment in these areas. He’s the inventor of a number of applied sciences together with basic technical contributions in these fields with 86 patents, 40 papers, and a graduate degree textbook, and recipient of the AT&T Science and Expertise Medal.
