Flow-based Packet Processing Framework on DPDK and VPP
summary
A flow-based packet framework built on DPDK and VPP implements a centralized flow table, NIC hardware offload, and a software fallback when NIC rules overflow, guaranteeing packet order and shared state across features. It supports IPv4/IPv6, bridge-domain and VRF awareness, policy-based routing and tunnel offload, and maps one flow to many NICs (and vice versa). Claimed performance shows viability for real deployments, with demonstrations handling 10,000 streams concurrently.
AI-written summary · about
chapters (19)
- 0:00Intro
- 0:33Acknowledgement
- 2:03Problem Statement
- 3:17Flow-based Solution
- 6:46Intel E810 NIC
- 7:53Flow Configuration
- 9:15HW Flow Offloading
- 10:53NICS & Flows
- 11:19BD/VRF Aware
- 12:13DPI Flow Processing
- 13:20TCP Connection Track
- 14:20Appl Database
- 15:18How Hyperscan Works
- 16:25Block Mode and TCP Segments Reassembly
- 18:03Stream Mode and TCP Segments Reassembly
- 19:05Identifying Layer 7 Applications
- 21:22Test Stream
- 21:53Test Result
- 23:05Key Takeaway
related talks
10mUsing VPP as Envoy's Network Stack
34mCoupling Eventdev Usage with Traffic Metering & Policing (QoS)
18mPower Savings & NFV: How to reconcile these 2 constraints ?
24mCan a User-Space Protocol Stack Balance High Performance and Generality? Our Answer: Both.
25meBPF Strengthens SR-IOV to Be Powerful
31mHigh-Performance Cloud Networking With Calico and Vec...
23mBuilding a network stack for optimal throughput / low-latency trade-offs
22mEiffel: Efficient and Flexible Software Packet Scheduling