High-level Programming for Application Networks
summary
AppNet is a framework that compiles high-level generalized match-action programs for RPC processing into platform-specific code (RPC library, colocated proxy, remote proxy) and uses symbolic abstraction and execution to prove semantic equivalence when reordering or relocating stateful elements. It implements shared state with strong or weak consistency, integrates with Kubernetes, and runs on three platforms. Common ANFs take 7–28 lines of code to express; optimizations cut RPC processing latency by up to 82% in evaluations including a hotel-reservation benchmark.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:48Service mesh challenges
- 4:19AppNet architecture
- 6:49Programming abstraction
- 8:16Compiler and semantics
- 11:26Usability and expressiveness
- 12:50Performance improvements
- 14:16App performance results
- 15:00Conclusion and Q&A
related talks
16mNetRPC: Enabling In-Network Computation in Remote Procedure Calls
47mDeep Programmability: A New Lens on Networking
21mTowards An Application Objective-Aware Network Interface
07mApplication Networking is Broken - Platform Engineering to the...
16mAutomatic Parallelization of Software Network Functions
18mAPKeep: Realtime Verification for Real Networks
01h 01mProgramming Abstractions for Orchestration of HPC Scientific Computing
48mCalling Functions Across Languages