KVM Latency Performance Tuning
summary
Implements a fastpass for virtual IPI via the KVM exit path that sends fixed single-target virtual IPIs before host interrupts and RCU, showing ~30% latency drop on an API benchmark and 22.3% on a KVM unit test; evaluates AMD SVM hardware acceleration for IPIs which gives ~3% hackbench gain but causes a 55.2% multicast IPI regression; adds a virtual TSC deadline fastpass with ~16.5% timer-latency reduction; introduces boosting preempted vCPUs and a new hypercall to target delivered IPIs, delivering 3.4%–48.3% improvements across 1–3 VMs in their tests.
AI-written summary · about
chapters (7)
- 0:00Start
- 1:15Virtual IPI Fastpath
- 2:47IPI AMD hardware acceleration
- 3:55Virtual TSC-Deadline Timer Fastpath
- 7:03Yield To IPI Target
- 7:32Boost preempted VCPU and Yield
- 8:27Reference
related talks
26mLessons Learned: Optimizing KVM Performance for EHR Systems
45mDynamic vCPU priority boosting in KVM for latency sensitive workloads
39mEvaluation of PREEMPT_RT in Virtualized Environments
25m"Enlightening" KVM: Hyper-V emulation
29mA Hidden Cost of Virtualization When Scaling Multicore Applications
31mKVM Nested Virtualization: Workings and Usecases
27mMitigating speculative execution attacks with ASI - follow up
11mUserspace Bypass: Accelerating Syscall-intensive Applications