«

The virtualisation performance varies a lot depending on the approach
taken. Here are some oldish benchmarks:

http://www.cl.cam.ac.uk/Research/SRG...rformance.html

They are Linux only but they compare some different approaches: Native
vs Xen vs Vmware 3.2 (old version due to newer licenses preventing
publication), and User Mode Linux.

As you can see Xen (the hypervisor approach) doesn't drop much below
native performance - whereas Vmware and UML really suffer on some
benchmarks. This gives a somewhat rough idea of what should be possible
for Virtual Server and VMware once adapted for the newer hardware.

I for one welcome our new virtualisation overlords

Hello AD.,

> I for one welcome our new virtualisation overlords

lol..

i like the whole idea of virtualization, go into any moderatly sized data
center and find out why. its absolutely crazy, racks and racks of servers
crunching away (well crunching away 10% of the time).

if you can load up some of your 'lower end' applications into some virtualized
enviroments you can save some big bucks. save some big bucks for some metal
that runs your big apps.

that being said, how much stuff does the likes of a dedicated database server
have to do before all its cpu's are peaked at 100% could you virtualize some
big ass databases ?

----------------
Steven H

the madGeek

I reckon there are still some advantages even if you only have one
virtual machine running on a physical server.

That virtual server is now (somewhat*) decoupled from its hardware. You
can shut it down and bring it back up on different hardware without
much bother.

* depending on the approach taken.

All sorts of cool possibilities. It has lots of implications for stuff
like migrations, disaster recovery, or even juggling virtual servers
around to balance out and maximise available hardware resources.

eg on Xen (I'm not up with the latest features of the other systems),
if you use a cluster aware network filesystem for the virtual servers
data, a running virtual server can be transferred from one host machine
to another on the fly without it even realising.

Hello AD.,

> a running virtual server can be transferred from one host
> machine to another on the fly without it even realising.

**** thats cool..

what if ...

all the diffrent 'chunks' of an operating system are each compartmentalized
in their own vm ?

the disk i/o, network stack, user i/o (keyboard / mouse), kernel, and user
enviroment could all be in their own vm. it could mean that you could replace
the kernel without a system reboot - just pause the vm's load the new kernel,
resume the vm's

that would be ideal for mission critical (run a nuclear power plant) type
software

of course you will have the communication issues involved, sending messages
from one vm to the other but that could be gotten around by using some sort
of enhanced tcp.

----------------
Steven H

the madGeek

> I reckon there are still some advantages even if you only have one
> virtual machine running on a physical server.
>
> That virtual server is now (somewhat*) decoupled from its hardware.
> You can shut it down and bring it back up on different hardware
> without much bother.
>
> * depending on the approach taken.
>
> All sorts of cool possibilities. It has lots of implications for stuff
> like migrations, disaster recovery, or even juggling virtual servers
> around to balance out and maximise available hardware resources.
>
> eg on Xen (I'm not up with the latest features of the other systems),
> if you use a cluster aware network filesystem for the virtual servers
> data, a running virtual server can be transferred from one host
> machine to another on the fly without it even realising.
>

> **** thats cool..

Yeah, although I've never tried it myself

http://www.cl.cam.ac.uk/Research/SRG...00000000000000

Both servers need to be on the same L2 subnet (the virtual MAC and IP
addresses transfer too), and both need to access the vitrual servers
filesystems, but it apparently happens with only 60-300ms of downtime.

--
Cheers
Anton