The RCU-Reader Preemption Problem in VMs

When synchronization primitives such as locking and read-copy update (RCU) execute within virtual machines (VMs), preemption can cause multi-second latency spikes, increasing peak memory footprint and fragmentation inside VMs, which in turn may trigger swapping or VM ballooning. The resulting CPU utilization and memory footprint increases can negate the server-consolidation benefits of virtualization. Although preemption of lock holders in VMs has been well-studied, the corresponding solutions do not apply to RCU due to its exceedingly lightweight read-side primitives.

This paper presents the first evaluation of RCU-reader preemption in a virtualized environment. Our evaluation shows 50% increase in the peak memory footprint and 155% increase in fragmentation for a microbenchmark, 23.71% increase in average kernel CPU utilization, 2.9× increase in the CPU time to compute a grace period and 2.18× increase in the average grace period duration for the Postmark benchmark.