Freedom from memory interference
summary
Demonstrates that memory contention on an NVIDIA AGX causes up to ~10x latency blowups under attacker traffic and that combining cache partitioning with colouring (implemented in Zen hypervisor on a Zynq board) and bandwidth partitioning with MemGuard (in Jailhouse) restores timing: colouring removes the interrupt-dispatch outliers, and MemGuard reduces benchmark slowdowns from 2–4x back to near nominal. Minerva Systems supplies open patches and runtime configuration; experiments report concrete latency distributions and working-set sizes.
AI-written summary · about
chapters (9)
- 0:00SOAFEE - Freedom from memory interference
- 1:03Next-generation automotive architecture
- 2:00Where is this relevant for SOAFEE?
- 2:16Contention on memory hierarchy
- 4:07Manage memory hierarchy contention
- 4:57Separation of Responsibility
- 6:46Real-Time TT SOAFEE Architecture TSG
- 7:14Cache partitioning with colouring
- 10:28Bandwidth partitioning with MemGuard
related talks
11mUnderstanding and Tackling the Hidden Memory Latency for Edge-based Heterogeneous
01h 04mWe Live in Interesting Times
46mBuilding Mixed Criticality Linux Systems with the Jailhouse Hypervisor
22mOld School cache coherence-a snooping system for a multicore processor based on CVA6
22mPreview: Hardware and Offloads
18mHarvesting Idle Memory for Application-managed Soft State with Midas
35mZephyr OS Memory Protection
18mMemSpec: Memory-Aware Runtime for Adaptive Draft Scheduling in Speculative Decoding on(…)