Mixed-Criticality Pod Scheduling: Linux cgroups v2, Core Pinning, and RT Priorities

Preventing non-critical entertainment apps from degrading steering and braking safety microservices using hierarchical cgroups v2 bandwidth controllers.

cgroups v2 Unified Hierarchy vs Legacy v1 Dual-Stack Vulnerabilities

Legacy Linux cgroups v1 utilized independent resource trees, creating dangerous loopholes: a process could be throttled on CPU but simultaneously execute unbounded buffered disk writes that starved memory. cgroups v2 enforces a single unified hierarchy where memory and I/O controllers cooperate atomically, guaranteeing that an application cannot bypass resource budgets.

Real-Time SCHED_FIFO Priority Bands for ASIL-D Actuator Microservices

While general-purpose microservices run under the standard Completely Fair Scheduler (SCHED_OTHER), critical pods (such as torque distribution and braking control) are granted POSIX SCHED_FIFO with static priorities above 90. The kernel preempts any standard user-space thread within 12 microseconds, guaranteeing that safety control loops execute deterministically.

cgroups v2 memory.max and memory.high enforcing instantaneous OOM termination for non-critical pods.
Static CPU core pinning for ISO 26262 ASIL-D safety pods with SCHED_FIFO real-time priority 99.
io.weight and io.max throttling preventing multi-gigabyte log dumps from starving safety sensor storage.
Kernel namespace sandboxing isolating IPC endpoints, PID trees, and network interfaces between domains.

Explore Domain Portfolio