HyperCom: Enabling High Performance and Composable Data Structures for Software Network Functions with eBPF

Bin Yang, Dian Shen, Hanlin Yang, Lunqi Zhao, Jianrui Liu, Jiantao Cheng, Beilun Wang

Abstract: The extended Berkeley Packet Filter (eBPF) has become a crucial component for modern network infrastructure, where the data structures, play a fundamental role. However, directly implementing these data structures with eBPF poses two main challenges: (1) unable to implement a full set of required functions due to limited programmability; (2) degraded performance caused by a restricted instruction set and runtime environment. This paper presents HyperCom, a framework that enables high-performance, comprehensive and flexible data structures with eBPF. At its core, HyperCom decouples one single data structure into two separate yet composable parts: base-complex but structurally stable across different network functions, and extension-simple yet functionally dynamic, capturing individual network functions, and applies right techniques to implement each of them. Specifically, HyperCom designs non-invasive and invasive extensions to support a flexible combination of base and extension. This approach achieves high performance by employing advanced hardware instructions for the base part, while providing flexibility by leveraging eBPF to fully implement the simplified extension part despite limited programmability. Furthermore, HyperCom leverages RUST to implement base to ensure safety. Evaluations on real-world network functions show that HyperCom-based ones outperform their eBPF-based counterparts by up to 2.5 ×in terms of packet processing rate, without compromising flexibility.

External IDs:dblp:conf/infocom/YangSYZLCW25