QUIC Performance Measurements
|Scientists:||Johannes Zirngibl, M.Sc., Benedikt Jaeger|
Research of the new QUIC protocol is a priority at our chair since early stages of the IETF draft. Our research focuses on the performance of QUIC on high-speed networks but also the protocol deployment on the Internet.
If you are an interested researcher or student who wants to cooperate with us or if you have any questions, feel free to contact us. Our QUIC interest group is for exchanging ongoing research, concepts, and measurement methodologies to evaluate the performance of QUIC and its deployment. We have collaborated with many students, supervised theses and presented results to the research community.
We conduct performance measurements of the QUIC protocol and different implementations on testbeds of our chair. All measurements are conducted on real hardware and focus on high-speed scenarios with links of 10Gbit/s or higher. We evaluate different libraries such as LSQUIC, Quiche and quic-go. Measurements are conducted in different scenarios and under different network conditions.
Previous and ongoing work on TCP at our chair allows us to compare new developments to existing, optimized stacks. Furthermore, the chair's focus on reproducible measurements and experiments provides us a basis for a detailed analysis of QUIC.
Therefore, we developed a framework to enable high-speed network measrurements on dedicated hardware servers. It is based on the QUIC Interop Runner and follows different principles: flexibility, portability and reproducibility. For more information we refer to our publication QUIC on the Highway: Evaluating Performance on High-rate Links . The framework is publicly available alongside all measurement configurations, and results shown in the paper, analysis scripts to parse the results, and Jupyter notebooks for visualization. We evaluate the influence of different system parameters and configurations and reach goodputs for default implementations of different QUIC libraries of up to 5 Gbit/s. The framework is presented at IFIP Networking 2023 and source code as well as raw data is provided via Github . Below you can find the general workflow used in our measurements.
Using this setup, we conducted measurements testing clients and servers against each other. Client and server run on dedicated hosts equipped with AMD EPYC 7543 processors and are connected via a 10G link. Then the client requests a file of size 8GB from the server via HTTP/3, which is then downloaded and the resulting goodput is computed using the duration of the download. The prevent packet drops by the operating system we increased the UDP receive buffer for the measurements.
Analyzing Existing Deployments
Together with GINO, we evaluate the deployment of QUIC on the Internet. We implemented tools to identify QUIC and HTTP/3 deployments shortly before the final release of RFC9000. The published QScanner allows to effectively scan QUIC capable targets, conduct a full IETF conform handshake and extract QUIC, TLS but also HTTP information.
Furthermore, we evaluated the configuration of deployed QUIC servers and shared parameters. Insights into deployed libraries, and parameters are combined with local measurements.
If this sounds interesting to you, feel free to contact us:
|2023-07-01||Johannes Zirngibl, Patrick Sattler, Georg Carle, “A First Look at SVCB and HTTPS DNS Resource Records in the Wild,” in International Workshop on Traffic Measurements for Cybersecurity 2023, Jul. 2023. [Pdf] [Bib]|
|2023-06-01||Benedikt Jaeger, Johannes Zirngibl, Marcel Kempf, Kevin Ploch, Georg Carle, “QUIC on the Highway: Evaluating Performance on High-Rate Links,” in International Federation for Information Processing (IFIP) Networking 2023 Conference (IFIP Networking 2023), Barcelona, Spain, Jun. 2023. [Preprint] [Rawdata] [Bib]|
|2021-11-01||Johannes Zirngibl, Philippe Buschmann, Patrick Sattler, Benedikt Jaeger, Juliane Aulbach, Georg Carle, “It’s over 9000: Analyzing early QUIC Deployments with the Standardization on the Horizon,” in Proceedings of the 2021 Internet Measurement Conference, New York, NY, USA, Nov. 2021. [Preprint] [Homepage] [Rawdata] [Recording] [DOI] [Bib]|
Finished student theses
|Simon Karan Guayana||Analyzing the Effect of Transport Parameters on QUIC’s Performance||BA||Johannes Zirngibl, Benedikt Jaeger||2022|
|Marcel Kempf||Analysis of Performance Limitations in QUIC Implementations||MA||Benedikt Jaeger, Johannes Zirngibl||2022|
|Florian Gebauer||Evaluating Different QUIC Scan Approaches||BA||Johannes Zirngibl, Patrick Sattler||2022|
|Mohammad Shaharyar Shaukat||Measuring the Impact of Transport Layer Protocols and Their Configuration on the Performance of Connections||MA||Simon Bauer, Patrick Sattler, Johannes Zirngibl||2022|
|Michael Kutter||Evaluation of Scalability and Limitations of HTTP/3||BA||Benedikt Jaeger, Johannes Zirngibl||2022|
|Kevin Ploch||QUIC Performance on 10G Links||BA||Benedikt Jaeger, Johannes Zirngibl||2022|
|Sebastian Voit||Bringing QUIC to High-speed Networks||MA||Benedikt Jaeger, Johannes Zirngibl||2021|
|Daniel Hegedüs||The First Year of QUIC v1 Deployment||BA||Johannes Zirngibl, Patrick Sattler, Benedikt Jaeger, Juliane Aulbach||2021|
|Philippe Buschmann||Analyzing Quic in the wild||MA||Johannes Zirngibl, Patrick Sattler, Benedikt Jaeger, Juliane Aulbach||2020|
|Marcel Mussner||In Depth Analysis of QUIC’s Lack of Kernel Optimizations||MA||Benedikt Jaeger, Johannes Zirngibl||2020|
|Lennart Keller||Packet Pacing with the QUIC Protocol||BA||Benedikt Jaeger, Johannes Zirngibl||2020|
|Marcel Kempf||Evaluation of the QUIC Spin Bit for RTT Estimation||BA||Benedikt Jaeger, Johannes Zirngibl||2019|
Open and running student theses
|Felix Christ||MASQUE-Proxying in User-Space||MA||Kilian Holzinger, Lion Steger||2023|
|Luca Otting||Improving QUIC with User Space Networking||BA||Kilian Holzinger, Benedikt Jaeger, Johannes Zirngibl||2023|
|Niklas Beck||Root Cause Analysis for Throughput Limitations of QUIC Connections||MA||Simon Bauer, Johannes Zirngibl||2023|
|Johannes Späth||QUIC Performance Improvements Using DPDK||MA||Johannes Zirngibl, Benedikt Jaeger, Kilian Holzinger||2023|
|Raphael Stadler||QUICkly Reaching Maximum Throughput: A Comparative Evaluation of QUIC Implementations||BA||Benedikt Jaeger, Johannes Zirngibl||2023|
|Nikolas Gauder||Performance Evaluation of Cryptography in QUIC||BA||Benedikt Jaeger, Johannes Zirngibl||2023|
|Moritz Buhl||QUIC Kernel: an In-Kernel Port and Socket Abstraction Layer||IDP||Johannes Zirngibl, Benedikt Jaeger, Kilian Holzinger||2023|
|reserviert||Forward Erasure Correction Coding in QUIC||BA, MA, GR, IDP||Kilian Holzinger||2023|
|Christoph Rotte||C++-based MASQUE-Proxying for Lower OSI-Layer Protocol Traffic||IDP||Lion Steger, Richard von Seck||2022|
|Christoph Probst||Rust-based MASQUE-Proxying for Lower OSI-Layer Protocol Traffic||MA||Lion Steger, Richard von Seck||2022|