The problem
How much scheduling work can a large machine afford before coordination becomes part of the workload?
What current research shows
Linux balances work across per-CPU run queues and scheduling domains while accounting for capacity and topology. The kernel documentation notes that the algorithmic cost of balancing and its impact on shared structures grows more than linearly with the number of CPUs being balanced.
Where the evidence stops
The exact cost is implementation- and topology-dependent. A large theoretical machine and a production host can have very different bottlenecks.
What Valen Systems is testing
Measure scheduler CPU time, lock or synchronization behavior, migrations, wakeup latency, and useful work as worker count and topology scale.