SUPR
DL for Network Trace Analysis
Dnr:

NAISS 2023/22-1277

Type:

NAISS Small Compute

Principal Investigator:

Johan Garcia

Affiliation:

Karlstads universitet

Start Date:

2023-12-21

End Date:

2025-01-01

Primary Classification:

10201: Computer Sciences

Webpage:

Allocation

Abstract

The project has two aims: Provide resources for competence development regarding using Deep Learning and Transformer models, primarily for analysis of traces from computer networks. In particular, wireless computer networks with time varying throughput characteristics are of interest, such as Starlink and 5G cellular networks. Starlink in particular is intersting since the radio resource allocation mechanism in essence is a black box. In our previous research [1], [2], [3] we have employed high-precision network measurements to infer the until now unknown Starlink physical layer rate, confirm frame timing, reallocation periods, and identify send burst characteristics. DL and transformer models are promising approaches for extracting further structural knowledge from the same Starlink time data set, and similar 5G data sets we have. From the obtained structural knowledge, models for predicting optimal queue management configuration and path-scheduling in multi-path contexts can be crafted. A second line of investigation concerns analysis of privacy attacks utilizing network traces, which considers the analysis of traffic similar to what is described in [4], and a planned extension of an already developed framework for DL privacy attack analysis/network fingerprinting [5] to the use of parallel resources. [1] "Multi-Timescale Evaluation of Starlink Throughput", Johan Garcia et al., Proc 1st ACM Workshop on LEO Networking and Communication, 2023 [2] "Fine-Grained Starlink Throughput Variation Examined with State-Transition Modeling", Johan Garcia et al. Accepted for publication in Proc 19th Wireless On-demand Network systems and Services Conference, 2024 [3] "Inferring Starlink Physical Layer Transmission Rates Through Receiver Packet Timestamps", Johan Garcia et al. Under review [4] "Splitting Hairs and Network Traces: Improved Attacks Against Traffic Splitting as a Website Fingerprinting Defense", Matthias Beckerle et al., Proc ACM SIGSAC Conference on Computer and Communications Security, 2022 [5] https://github.com/m-bec/Splitting-Hairs-and-Network-Traces