CountingStars: Low-overhead Network-wide Measurement in LEO Mega-constellation Networks

TL;DR

CountingStars is a novel measurement architecture for LEO satellite networks that significantly reduces memory usage and measurement errors by predicting network topology and using collision-free hash seeds.

Contribution

It introduces a digital twins system for topology prediction and a port aggregation data structure to address memory inflation and hash collisions in satellite network measurement.

Findings

Memory usage reduced by 70% on average

Measurement error reduced by 90% on average

Feasible FPGA implementation demonstrated

Abstract

The high mobility of satellites in Low Earth Orbit (LEO) mega-constellations induces a highly dynamic network topology, leading to many problems like frequent service disruptions. To mitigate this, Packet-based Load Balancing (PBLB) is employed. However, this paradigm shift introduces two critical challenges for network measurement stemming from the requirement for port-level granularity: memory inflation and severe hash collisions. To tackle these challenges, we propose CountingStars, a low-overhead network-wide measurement architecture. In the ground controller, CountingStars builds a digital twins system to accurately predict the future network topology. This allows ground controller to generate and distribute collision-free hash seeds to satellites in advance. On the satellite, we introduce a port aggregation data structure that decouples the unique flow identifier from its multi-port counter and updates it through efficient bit operations, solving the memory inflation caused by PBLB. Simulation results show that the memory usage of CountingStars is reduced by 70\% on average, and the relative error of measurement is reduced by 90\% on average. Implementation on FPGA shows its prospect to deploy in real system.