On Monday 10 June 2002 03:52 pm, Rick Troth wrote:
> What David Rock said.
> And I don't think VMware is at all to blame.
> I *suspect* that virtualization of Pentium is hard to do.
> zSeries has more than three decades of hardware and software
> adjusting for each other to do virtualization very efficiently.

The Intel chips virtualize in what's called "virtual 8086 mode". Basically
what this means is that most of the non-privileged instructions execute at
full native speed, but any privileged instruction (including all I/O) must
be trapped and software-emulated.

The "virtualization" term is a bit of a misnomer, because they aren't really
doing microcode the way S/390 does. Intel is able to select a subset of its
instructions that are "safe" to allow to execute natively even inside an
environment that is nominally restricted. For example, "ADD AX,BX" simply adds
together the contents of two registers; there aren't any memory management
issues, setting of processor control bits, etc., so this is an intrinsically
safe instruction. They do some trickery with memory addressing as well,
using the memory management features introduced originally in the 80386.

This is what can be done in *hardware* by Intel chips; the rest of VMWare is
all software. I'm actually very impressed with the performance of VMWare on
the machines where I've run it; if you have enough RAM, its "insertion loss"
is reasonable especially since most PCs have CPU cycles to burn anyway. But
it doesn't compete against z/VM, really, through no fault of either VMWare
or Intel. They are different tools meant for a different purpose.

Scott

--
-----------------------+------------------------------------------------------
Scott Courtney         | "I don't mind Microsoft making money. I mind them
[EMAIL PROTECTED]       | having a bad operating system."    -- Linus Torvalds
http://www.4th.com/    | ("The Rebel Code," NY Times, 21 February 1999)

Reply via email to