The QEMU and KVM developers' efforts towards a high-performance block layer started back in 2012, and are now seeing the light. In this presentation I will lead the audience through this multi-year journey.
I will explain the structure of the QEMU block layer, why it has traditionally preferred cooperative multitasking (coroutines) to threads, and how to reconcile this design with the high level of parallelism required by modern PCIe storage devices.
Speakers: Paolo Bonzini