Loading…

Memory Bandwidth Throttling to Maximise Performance and Reduce Power Consumption

With the ongoing evolution towards "telco cloud", networks increasingly rely upon a continuum of compute-based infrastructure to host Virtualised and Containerised Network Functions (VNFs/CNFs). A key challenge that remains is how to apply sufficient platform tuning to achieve the requisit...

Full description

Saved in:
Bibliographic Details
Main Authors: Veitch, Paul, MacNamara, Chris, Browne, John J
Format: Conference Proceeding
Language:English
Subjects:
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:With the ongoing evolution towards "telco cloud", networks increasingly rely upon a continuum of compute-based infrastructure to host Virtualised and Containerised Network Functions (VNFs/CNFs). A key challenge that remains is how to apply sufficient platform tuning to achieve the requisite performance of the myriad network functions hosted on telco cloud infrastructure, including those associated with existing fixed broadband and mobile (4G and 5G) services. Performance targets will become more stringent with the advent of 6G, and the cloud continuum will extend deeper to the edge of the network. While achieving performance targets, the energy efficiency of the compute infrastructure should also be optimised. Many recent works on telco cloud energy efficiency have focused on processor-oriented power management and resource tuning, as this is a major contributor to server power consumption. This paper switches attention to the memory power consumption of servers typically used for telco cloud workloads and demonstrates the possibility of maximising performance of high priority workloads in the presence of a competing "Noisy Neighbour" workload, while achieving a non-negligible saving in power consumption attributed to system RAM (Random-Access Memory). The best-case scenario of combined platform tuning achieves a 45% improvement in throughput performance of a User Plane Function (UPF) workload, with a corresponding reduction of 34% in memory power consumption.
ISSN:2472-8144
DOI:10.1109/ICIN60470.2024.10494488