What Happens When 4096 Cores All Do synchronize_rcu_expedited()?

Paul McKenney
In order to operate efficiently on very large systems, algorithms must avoid high levels of contention on global locks, global counters, and pretty much all other global mechanisms. So what is a parallel algorithm to do? My usual advice might be expected to be "Use RCU!", but that advice is not always helpful within an RCU implementation.

This presentation will illustrate some useful parallel techniques, using expedited RCU grace periods as a vehicle for motivating these techniques and for demonstrating their effectiveness.