The problem
When should two logical CPUs share a physical core, and when should the system keep them apart?
What current research shows
Linux core scheduling groups tasks so incompatible tasks don’t run simultaneously on one core. The documentation also notes that coordination can perform worse than disabling SMT when runnable demand is low enough, because a sibling may be forced idle.
Where the evidence stops
Core scheduling reduces a defined class of co-scheduling risk; it doesn’t make every microarchitectural channel disappear, and its performance cost depends on workload mix.
What Valen Systems is testing
Compare shared-core throughput, tail latency, sibling idle time, and security policy outcomes under explicit task-group assignments.