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

[Sam Vilain]

My current hazy picture is; and no, I don't have any evidence ;-).

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.

Pure userland full VM solutions:
UML (standard) - 50% or less than host speed (I gather)

Hybrid solutions:
UML (``skas'' mode) - 70% of host speed (apparently)
Plex86+Bochs - something like 90% of host speed for simple ``tight loop'' 
tests
Plex86+Linux/HAL - potentially a little bit more than the above (?)

`chroot' based solutions:
FreeVSD/ProVSD - 99% of host speed (no host patch required, no true `root' 
in virtualised servers)
Vserver - 99% of host speed (requires host patch, true `root' in 
virtualised servers)

Of course, with the chroot based solutions, everything runs within the same 
kernel.  Calling it 99% of host speed is really a gross 
oversimplification; issues such as the extra memory and filesystem areas 
required by all the running servers will probably reduce that somewhat; 
but this is the same for all the solutions.

Also, different solutions have different levels of stability and vendor 
support.  For instance, if you're an ISP wanting to sell ``Virtual 
Dedicated Servers'' to clueless markets, ProVSD is the most turn-key 
solution.  Vserver based solutions are a bit more work, but IMHO worth it 
in the long run.

Currently the plex86 based solutions are far from actually achieving the 
above performance estimates for real world cases; but they should approach 
those over time.  So they can hardly be considered ready for production 
use.




On Thu, 20 Feb 2003 10:05, Shai Fultheim wrote:
> All,
>
>
>
> Has anyone done any performance test or comparison between the different
> VM products around?
>
> What should be the expected performance loss in case of VMware or Plex
> or VirtualPC??
>
> What are the numbers for bochs??
>
>
>
> Any info will be appreciated.
>
>
>
> _________________________
> Shai Fultheim

-- 
Sam Vilain, address@hidden

  Time is an illusion perpetrated by the manufacturers of space.
GRAFFITI