M.Sc./B.Sc. Topic: Pushing the Kernel Beyond its Limits: QUIC at 100 Gbps (without AI) ⚡️

QUIC is a relatively young transport protocol that can be used as an alternative to the established TCP. While TCP was traditionally used as the underlying transport for HTTP, version 3 of HTTP (H3 or HTTP/3) uses QUIC.

Researchers evaluated the performance impact of QUIC on HTTP by comparing H2 and H3 deployments. Shreedhar et al. found that H3 yields lower page load times than H2 but also an increased CPU load and lower bulk download durations [1]. Zhang et al. looked at the receiver-side performance of QUIC in more detail and found that the computational demands caused severe performance deficiencies compared to TCP [2]. Jäger et al. compared different QUIC implementations and found big performance differences between them [3]. Improving the throughput of QUIC can for example have an effect on throughput-intensive web applications such as video streaming, and on the number of clients handled by a server.

The purpose of this thesis is to evaluate the impact of different implementation techniques on QUIC’s performance. The work should look at optimizing the performance of a single QUIC sender/receiver on the one hand, and on improving the performance of a server handling multiple concurrent connections on the other. Promising technologies are Generic Receive Offload (GRO) and io_{_}uring that could be evaluated against epoll. Further investigations could look into parallelizing QUIC’s send or receive paths.

To successfully work on this topic, you need to have a profound interest in working with the Linux networking interfaces and good knowledge about transport protocols (TCP and ideally also QUIC).

The QUIC implementation that you would work on is quiche written in Rust. Existing experience with Rust is a plus but not required if you have a strong programming background. Other candidate QUIC implemenations are quic-go (Golang) and picoquic (C).

If this thesis proposal sparked your interest, please get in touch with Hendrik Cech (hendrik.cech@tum.de). Please briefly introduce yourself, state what you find most interesting about this topic, and include a TUMonline grade report.

References

[1] T. Shreedhar, R. Panda, S. Podanev, and V. Bajpai, “Evaluating QUIC Performance Over Web, Cloud Storage, and Video Workloads,” IEEE Transactions on Network and Service Management, vol. 19, no. 2, pp. 1366–1381, Jun. 2022, doi: 10.1109/TNSM.2021.3134562.

[2] X. Zhang et al., “QUIC is not Quick Enough over Fast Internet,” 2023. [Online]. Available: api.semanticscholar.org/CorpusID:264145954

[3] B. Jaeger, J. Zirngibl, M. Kempf, K. Ploch, and G. Carle, “QUIC on the Highway: Evaluating Performance on High-rate Links,” in International Federation for Information Processing (IFIP) Networking 2023 Conference (IFIP Networking 2023), 2023.