Interview with Andrew Tanenbaum (LinuxFr.org)

I'm actually pleased and curious about the idea that the equivalent of virtual memory can now be implemented in software, algorithmically. This is something that I've been hoping that would happen because there's a promise that running everything inside dynamically compiling kernel makes all JIT-based languages more efficient to run.

As an example, the current process-based isolation tends to mean that every process using JIT must invent its own versions of dynamically compiled bytecode of the same libraries they might all be using, so clearly from architectural point of view they stand at some disadvantage to ahead-of-time compilers where shared libraries exist. This advantage, in turn, is realized by combination of hardware and software support.

Yet I think overall implementing virtual memory in software is where I would want to be, because the managed world seems nicer to program in.