6.3 Solutions for energy consumption

28.3103GPPEnergy efficiency of 5GManagement and orchestrationRelease 18TS

6.3.1 Solution for energy consumption of PNFs

TS 28.552 [15] clause 5.1.1.19 defines measurements for the Energy Consumption (EC) of Physical Network Functions (PNF), associated to corresponding ManagedElement IOC instances.

The method for collecting these measurements is described in ETSI ES 202 336-12 [4].

6.3.2 Solution for energy consumption of VNF/VNFCs

6.3.2.1 Introduction

In case of Network Functions (NF) composed of Virtualized Network Functions (VNF) running on a Network Function Virtualization Infrastructure (NFVI), it is expected to be able to measure the energy consumption of each VNF separately. However, in a NFVI, the finest grain at which Energy Consumption can be measured is the NFVI Node, making it impossible to measure the energy consumed by each and every VNF separately given that a) a VNF can run on more than one NFVI node and b) a NFVI node can support more than one VNF. Therefore, this clause describes a solution for estimating the energy consumption of VNFs.

ETSI GR NFV-IFA 015 [19] states that:

– a VNF is composed of 1-to-many VNF Component(s) (VNFC) – see diagram below.

– a VNFC runs over a single VirtualisationContainer – see diagram below.

Figure 3.2.1-1: VNF-VNFC-Virtualisation Container relationship

where a Virtualisation Container is defined in ETSI GR NFV 003 [20] as follows:

"

partition of a compute node that provides an isolated virtualised computation environment.

NOTE: Examples of virtualisation container includes virtual machine and OS container.

".

Hence, a Virtualisation Container runs on a single NFVI Compute Node. A NFVI Compute Node may support 1-to-many Virtualisation Container(s).

To estimate the Energy Consumption of VNF / VNFCs, it is assumed that:

– Pre-condition #1: there exists a Management Function (MF) in charge of estimating the energy consumption of the VNFs.

– Pre-condition #2: this MF knows on which NFVI node(s), the VNF/VNFC instances run;

– Pre-condition #3: NFVI nodes are equipped with embedded or external sensors (see ETSI ES 202 336-12).

6.3.2.2 Solution for VM-based VNF/VNFCs

6.3.2.2.1 Solution based on vCPU usage of virtual compute resources

The procedure for estimating the energy consumption of VNF/VNFCs based on the vCPU usage of underlying virtual compute resources is as follows:

1. The MF in charge of estimating the energy consumption of VNFs collects Power, Energy and Environmental (PEE) measurements from NFVI nodes (see clause 6.3.1), during a given period of time. The procedure described here is independent from whether the NFVI nodes are equipped with embedded sensors or external sensors;

2. The MF subscribes to PM notifications towards the VNFM, so as to receive notifications about the vCPU mean usage of selected VNF/VNFC instances (see ETSI GS NFV IFA 008 [14] clause 7.4.4) for a given period of time (same observation period as in 1);

3. The MF requests the VNFM to create a PM job to collect the vCPU usage of selected VNF/VNFC instances (see ETSI GS NFV IFA 008 [14] clause 7.4.2);

4. The VNFM subscribes to PM notifications towards the VIM, so as to receive notifications about the vCPU usage of the virtual compute instances on which each VNF/VNFC instance runs (see ETSI GS NFV IFA 006 [20] clause 7.7.5);

5. The VNFM requests the VIM to create a PM job to collect the vCPU usage of the virtual compute instances on which each VNF/VNFC instance runs and whose IDs are provided as input parameters of the CreatePMJob request (see ETSI GS NFV IFA 006 [20] clause 7.7.2);

6. The VIM gets, at pre-defined intervals, the process utilization compute metric values from all CPU Cores of the NFVI (see ETSI NFV TST 008 [14] – clause 6.6). Whether the VIM gets this data in pull mode or in push mode is out of scope of the present document;

7. The VIM aggregates them per virtual compute resource and calculates their arithmetic mean per virtual compute resource; this per virtual compute resource arithmetic mean of process utilization compute metric values is called VCpuUsageMean (see ETSI GS NFV IFA 027 clause 7.1.2);

8. The VIM notifies the VNFM about VCpuUsageMean measurement(s) for the virtual compute instance(s) (see ETSI GS NFV IFA 006 [20] clause 7.7.6);

9. The VNFM maps the received VCpuUsageMean measurement(s) from virtual compute instances to the VNF/VNFC instance(s);

10. The VNFM generates the measurement for the subject VNF/VNFC instances by assigning the value of the multiple VCpuUsageMean measurements received (see ETSI GS NFV IFA 027 [18] clause 7.2.2);

11. The VNFM notifies the Management Function in charge of estimating the 5GC NF EC, about the average VCpuUsageMean of each virtual compute instance used by the VNF/VNFC instance(s) which constitute the NF (see ETSI GS NFV IFA 008 [14] clause 7.4.5);

12. NF energy consumption can be now estimated as follows:

– The energy consumed by the NF is the sum of the energy consumed by all its constituent VNF/VNFC instances.

– For each VNF/VNFC instance, its estimated energy consumption is a proportion of the NFVI node energy consumption on which it runs.

– This proportion is equal to the vCPU mean usage of the VNF/VNFC instance relatively to the sum of the vCPU mean usage of all VNF/VNFC instances running on the same NFVI node.

Annex A (informative):
Plant UML source code