Re: [Plex86-devel] Re: [Plex86-developers] Performance characteristics

[Drew Northup N1XIM]

Ok, let me take the chance to clear something up: Emulators, "Virutal
PC's," Software/Hardware Abstraction Layer Virtual Machines, and true
Hardware Virtualization Virtual Machines--what are they, and how are
they different?

The first class, Emulators, takes something that you don't have and
pretends that you do have it.  It may be some chunk of hardware or some
object of software that you don't have--but with the emulator you don't
need it.  Bochs is the perfect example of a clean and simply-directed
emulator (a very good thing indeed).

The second class is a type of product often confused with a real
engineering construct--the "Virtual PC."  Bochs, VMWare, and Plex86 (and
many other things...) are all Virtual PC Products.

Next there are Software/Hardware Abstraction Layer Virtual Machines. 
Things like User Mode Linux (here, UML) and chroot environments are
items of this sort.  So is the TCP/IP suite of protocols.  Let's not
forget Ethernet either.  Both of those are part of the larger OSI
network machine model.  The actual ISA of a chip is an abstraction
layer--a virtual machine in this sense.  I hope that my point here is
clear.  This is the most traditional idea of a virtual machine.

Finally we come to the true Hardware Virtualization Virtual Machines. 
Plex86 is (in all of its somewhat working forms) a good example of
this.  When used to co-simulate with Bochs, however, Plex86 becomes more
of something from the third class of items that I previously talked
about.  Another noteworthy example is OS/370 from our buddies at the
Itty-Bitty-Machines company.  In OS/370 each virtual machine gets a
complete copy of the hardware--except that it is a virtual one.  Read
"Exokernel."  

This is something that I'd like to see done with Plex86 some time.  In
fact, in little itty-bitty parts I'm working to that end.  Kevin has his
goals and ideas, I have mine.  There's nothing wrong with this, and I
intend to help Kevin out if I possibly can (mostly crash-testing his
code, since I'm still working on deciphering the last batch).

So there it is.  I hope that clears a few things up.

--Drew Northup, N1XIM

On Thu, 2003-02-20 at 01:06, niraj tolia wrote:
> 
> 
> >From: Sam Vilain <address@hidden>
> 
> [snip]
> 
> >Full x86 VM solutions:
> >VMWare, old Plex86, VirtualPC - qualitatively agreed are all extremely slow
> >:-)  This is because they are spending a lot of time performing
> >modifications of code segments to work around the x86 virtualisation
> >inadequacies.
> >
> 
> Actually VMWare gets very close to raw performance for operations that do 
> not need to be trapped by the VM. In other words, I have observed VMWare to 
> be quite fast for certain applications. For example, VMWare is fast enough 
> to saturate 100 Mbit connections.
> 
> VirtualPC on the other hand is an emulator (think bochs) unlike both the old 
> Plex86 and VMWare so you can't make a fair comparision.
> 
> Niraj
> 
> http://www.andrew.cmu.edu/~ntolia
> 
> _________________________________________________________________
> The new MSN 8: advanced junk mail protection and 2 months FREE* 
> http://join.msn.com/?page=features/junkmail
> 
> 
> 
> _______________________________________________
> Plex86-devel mailing list
> address@hidden
> http://mail.nongnu.org/mailman/listinfo/plex86-devel