Except that from a locking latency perspective, virtualization isn't a great choice. As Linus explains, guest operating systems runs on virtualized CPUs which c

Except that from a locking latency perspective, virtualization isn't a great choice. As Linus explains, guest operating systems runs on virtualized CPUs which can be de-scheduled while holding a spin lock. This is very much like what happened with the original benchmark in Malte's blog post, except that the application and operating system are now replaced with guest operating systems and hypervisors, respectively. AFAIK, the Linux kernel makes heavy use of spinlocks, making this a problem for people who care a lot about scheduling performance. This isn't to mean that people should avoid virtualization, though. There are likely to be many other bottlenecks to consider before most people have to consider these kind of stuff.