VMware Workstation: Bringing Virtualization to the x86 Architecture (2012) [pdf]

Impressive deep dive, this classic ASPLOS paper shows that early hardware virtualization support (like Intel VT-x) didn’t outperform VMware's binary-translated software VMM due to high VM‑exit overheads and rigid models/

The obvious takeaway? Flexible software optimizations often beat hardware if exits are too heavy or inflexible. Makes me wonder: with modern nested virtualization and microarchitectural improvements, are we finally seeing hardware VMMs that consistently match or exceed software VMMs?

We have also worked out how to vmexit less, eg more effective ways to do IO.

Hardware virtualization is cheating by using unsecure enhancements. Like 90% of existing CPUs have security vulnerabilities, that must be patched in OS.

We can have this discussion when hardware gets a few years without major security flaw!