There's an interesting set of comments by keithmadams on Twitter https://twitter.com/keithmadams/status/1577483553281310721 https://twitter.com/keithmadams/stat

There's an interesting set of comments by keithmadams on Twitter https://twitter.com/keithmadams/status/1577483553281310721 https://twitter.com/keithmadams/status/1577483553281310721, regarding the accuracy of this: > This article's history of "virtualization" begins with paravirt in Zen in 2004. Then, after hardware support emerged, v12n started working with with unmodified OSes. No part of this is true. VMware was running binary OSes without hardware support in 1999. > We did this by JIT'ing unrestricted x86 to a safe x86 subset. User-level code, with few exceptions, was directly executed, but kernel-level code ran through a little JIT. We described it in https://www.vmware.com/pdf/asplos235_adams.pdf https://www.vmware.com/pdf/asplos235_adams.pdf > It performs pretty well, as described above. The only problem is that tiny JITs that run in kernel mode and use x86 segmentation to hide themselves in the top 4MB of linear memory are ... hard to write. And VMware was making a lot of money. > Zen invented paravirt; AMD invented SVM; and Intel invented VT in parallel all around 2003-2004, not to make virtualization on x86 practical, but to make it easy enough to disrupt VMware's emerging monopoly. > As a VC focused on companies exploiting a tech advantage, VMware remains my favorite case study. For a time, they could do something that was undeniably valuable, that nobody else could do. And, moats like that are always time-limited; their value invites their destruction.