On VMI

Here, VMI stands for the "Virtual Machine Interface," not Vendor Managed Inventory, nor Virginia Militia Institute. :o

As I wrote several times recently, I'm working on virtualization these days.
At work, I did some benchmarks and saw not negligible CPU cost using VMware ESX Server.
But, I've been thinking paravirtualization using VMI would improve the situation greatly.

Since then, I got a draft version of VMI and tried to run VMI aware Linux kernels on top of that.
Although I'm still struggling doing so at work, today finally I saw VMI ROM detection message using VMware Workstation 6.0 and Fedora Core 9 (i386).
Hoho, what kind of benchmark should I do? :)

BTW, reading the VMI 2.5 specification, I thought why they didn't include a special I/F intended for SMP configuration, something like a 'SpinLock Ops.' As early implementation of Virtual Iron did, that can be a good clue for constructing SMP virtual machine.
I heard that a VMware guy mentioned at VMWorld 2007 that they made a design choice forcing to schedule all virtual CPUs of a guest simultaneously mainly because of spinlock consideration. But, I think the spinlock ops could be a clue for them.