Why I reimplemented LVM (with worse guarantees)
Summary
The article discusses reimplementing LVM for microVM-ready workloads on bare-metal hypervisors, arguing that the standard LVM approach is too slow due to locking and guarantees. It describes a layered approach that uses the kernel's device-mapper for the data plane while rebuilding the control plane in-process for speed and crash resilience, along with the tradeoffs and potential future directions.