ETHOS: Demystifying Performance, Energy, and Computational Efficiency in Virtualized 5G O-RAN Networks
Nov 4, 2025·
,·
0 min read
Zongshen Wu
Rahman Doost-Mohammady
Ashutosh Sabharwal
Abstract
Virtualized O-RAN promises flexibility and operator-defined behavior,
but the joint cost — in throughput, latency, energy, and compute — of
running 5G RAN functions on commodity servers is poorly understood.
ETHOS is a measurement and characterization study that disentangles
these costs across deployment configurations (CPU pinning, NUMA
layout, accelerator usage, slice mix). We report the dominant
bottlenecks, the energy/throughput Pareto frontier of common
configurations, and design implications for ML-driven RAN testing
and optimization.
Type
Publication
In Proceedings of the ACM Workshop on Wireless Network Testbeds, Experimental Evaluation & Characterization (WiNTECH), 97–104
Authors
Authors
Rahman Doost-Mohammady
(he/him)
Assistant Research Professor
I work at the intersection of computer systems and wireless
networking — building programmable, intelligent infrastructure for
5G/6G and Open RAN. A common thread across my projects is algorithm
and system design across FPGA, GPU, and CPU — spanning real-time
massive MIMO baseband on commodity servers, many-antenna hardware
platforms, and learning-driven schedulers. I am currently focused on
virtualized Open RAN systems and ML-enabled RAN software design.
Authors