Co-opting Linux Processes for High-Performance Network Simulation
summary
Phantom is a discrete‑event simulator that directly executes unmodified Linux processes via an LD_PRELOAD shim, traps noninterceptable syscalls with seccomp trapping, and uses shared-memory remap of app stack/heap plus semaphores to avoid copies. It runs one virtual host per worker thread with logical processors and CPU pinning to prevent oversubscription. Phantom reports up to 2.2× speed vs Shadow, 3.4× vs NS‑3 and 43× vs gRaIL in P2P benchmarks, comparable performance to Shadow on Tor workloads and ~90% lower memory than uni‑process plugin approaches.
AI-written summary · about
chapters (14)
- 0:00Intro
- 1:34Requirements for Large Distributed System Experimentation
- 2:23Problems with Traditional Approaches
- 3:52Hybrid Architectures and Challenges
- 4:17Limitations of Hybrid Architectures (1)
- 6:52Design Overview
- 7:38Parallel Worker Threads
- 10:15Direct Execution
- 11:19Syscall Interposition
- 14:14Syscall Emulation
- 15:29Inter-Process Communication
- 17:22Process Control
- 19:16Evaluation: Large P2P Benchmarks
- 20:12Evaluation: Large Tor Networks
related talks
47mSpeedrun through Splicing Sockets with Sockmap
15mREAL: Emulating Control Plane at Simulator’s Cost
31mA Practical, Research‑Backed Introduction To Cloud‑Native Kernel‑Bypass Networking
20mNetKernel: Making Network Stack Part of the Virtualized Infrastructure
37mEfficient and Flexible Virtual Machine Networking Through eBPF
21mFine-Grained Isolation for Scalable, Dynamic, Multi-tenant Edge Clouds
25mShenango: Achieving High CPU Efficiency for Latency-sensitive Datacenter Workloads
34mCan the Linux networking stack be used with very high speed applications?