CS Events

PhD Defense Accelerating software packet processing for high-speed networks
Export Event Save iCal Send to Google Calendar Send to Yahoo Calendar Send to Outlook Live Send to Microsoft Outlook Save iCal Preserve formatting in description (only supported in some calendar applications) Close Thursday, August 29, 2024, 01:00pm - 03:00pm

Speaker: Qiongwen Xu

Location : CoRE 305

Committee: 

Prof. Srinivas Narayana (advisor/chair)

 Prof. Richard Martin

Prof. He Zhu

Prof. Mina Tahmasbi Arashloo (external) 

Event Type: PhD Defense

Abstract: The rapid growth of emerging applications, such as virtual reality (VR), deep neural network (DNN) model training and inference, and high-resolution video streaming, has significantly increased the demand for high-throughput and low-latency networks, which are often bottlenecked by packet processing. Software packet processing, such as the kernel network stack, is widely deployed and thus worth accelerating. This thesis focuses on software packet processing with the goal of bridging the performance between the current packet processing mechanisms and the full potential of the underlying hardware. Two systems are presented: K2, a compiler that optimizes BPF bytecode (i.e., a packet-processing program that can be attached in the kernel network stack) to enhance packet processing performance while ensuring formal correctness and safety (unsafe programs will be rejected by the BPF verifier and cannot be installed into the kernel), and SCR (state-compute replication), a principle for scaling throughput of stateful packet processing across multiple cores using replication. Experimental results demonstrate that K2 produces code with 6–26% reduced size, 1.36%–55.03% lower average packet-processing latency, and 0–4.75% higher throughput (packets per second per core) relative to the best clang-compiled program, across benchmarks drawn from Cilium, Facebook, and the Linux kernel. SCR enables linear scaling of packet-processing throughput with the number of cores, independent of flow size distributions.

: 

Contact  Prof. Srinivas Narayana

Zoom:
zoom link: https://rutgers.zoom.us/j/97049692390?pwd=LVGvtonLaHNtljjwdsDChpuxaGaNMr.1
Meeting ID: 970 4969 2390
Password: 803483