What are the practical (or real) advantages of Intel VT and TxTtechnology (Q8200 vs Q9300) ?

Robert Myers wrote:
> I had a really hard time visualizing a VM until I actually used one.
> Now things are much more clear, and, while I still don't understand
> the virtue of running 100 Linux servers on a mainframe, I can see some
> definite advantages for me:
>
> I can run Linux as a guest on Windows and have access to both
> environments at the same time.
>
> I'm assuming that it's the VT-x technology that allows me to run the
> virtual Linux box with it's own IP address. My router sees two
> distinct computers, in spite of the fact that I have but one ethernet
> card. I can ssh into the guest Linux box from the host box and use
> sftp to move files back and forth between the two machines.

The virtualized network interfaces actually predates the virtualized
processor environments by many years, and it's not really achieved
through the virtual processor mechanisms, nor is processor
virtualization necessary. It's simply a matter of changing the MAC ID of
the packets. Usually the MAC ID is whatever is built into the firmware
of the network adapter, or sometimes even into the BIOS firmware of the
host machine. But this identifier is interceptable and changeable.

Prior to processor virtualization, I used to run Sun machines with
multiple websites, each website was assigned its own virtual network
interface and was thus easily routable by MAC ID alone.

Yousuf Khan

On Aug 7, 5:13*pm, Yousuf Khan <bbb...@yahoo.com> wrote:

>
> The virtualized network interfaces actually predates the virtualized
> processor environments by many years, and it's not really achieved
> through the virtual processor mechanisms, nor is processor
> virtualization necessary. It's simply a matter of changing the MAC ID of
> the packets. Usually the MAC ID is whatever is built into the firmware
> of the network adapter, or sometimes even into the BIOS firmware of the
> host machine. But this identifier is interceptable and changeable.
>
> Prior to processor virtualization, I used to run Sun machines with
> multiple websites, each website was assigned its own virtual network
> interface and was thus easily routable by MAC ID alone.
>
I mentioned it mostly because the machine can talk to itself without
the intervention of a third party. For me, it's very convenient
because I don't have to worry so much about which environment
something is in.

Mostly, I was always aware of the privileged instruction issue and, no
matter whether it could be gotten around or not without VT-x, it
always sounded like it had to be baling wire and chewing gum, and I
had no interest in experimenting. That is to say, whether it's
supposed to work or not, I wouldn't bother with virtualization without
explicit hardware support for it because of the issues with
virtualization and x86. Good for vmware that they could work around
it; it wasn't worth any kind of risk for me. Now that it's well-
supported and can be had for free, it's pretty nifty.

Robert.

On Aug 8, 9:53*am, j...@cix.compulink.co.uk wrote:
> In article
> <9aeb629c-d9d1-43fa-9097-e069330c2...@r34g2000vba.googlegroups.com>,
>
> rbmyers...@gmail.com (Robert Myers) wrote:
> > ... while I still don't understand the virtue of running 100 Linux
> > servers on a mainframe
>
> One of the major advantages, from the PoV of IBM, and of the people in
> their customers who have built their careers round mainframes, is that
> it provides a new reason for having a mainframe.
>
> It can also take up less space and consume less power, and so on, but
> that depends on how hard your "Linux machines" are being worked, and
> what the other pressures on the organisation are.
>
It seems as if, by virtualizing Linux, you are giving up one of the
few genuine advantages of mainframes: they are usually not an
attractive target for hacking. If you virtualize Linux on them, the
only advantage you have left is that it's not x86, which advantage you
can get from other platforms (even and perhaps most compellingly
Itanium) at a much lower price.

Mainframes like to stay busy (translation: if your mainframes aren't
busy, you're burning big money), and aggregating lots of servers is
one plausible way of keeping any computer running servers, which spend
much of their time waiting, busy. I assume that the sales folks at
IBM have the pricing worked out (the same hardware is much cheaper if
it runs Linux) so they can make the case with a straight face, but
Itanium running HP-UX still sounds pretty attractive by comparison.
Not my side of the universe, anyway.

For people running lots of servers, the need for virtualization is
obvious and hardware support on x86 is a no-brainer if you're not
trying to protect Itanium, as Intel no longer is. For the rest of us,
it looks like a neat way to build sandboxes very cleanly.

> As an example, my employers have a whole lot of PCs, running Windows or
> Linux, running overnight each night testing the day's changes to
> CPU-bound software. We use the developer's desktops for this overnight,
> as well as lots of other machines. Notably, when a developer's desktop
> finishes its three-year depreciation life, it goes into the test farm,
> and does another 2-4 years there, depending on how fast it goes obsolete
> or breaks down.
>
> Some corporate IT management types have been trying to convince us to
> buy blade servers for the test farm. They claim it will save money, but
> we can't see how and they can't demonstrate it. We have enough space for
> the old machines. While we could cram the same CPU power into less space
> with blade servers, *we have no other use for the space unless we are
> allowed to hire some more people. Blade servers, like mainframes, are a
> more expensive way of getting the same CPU power than ordinary desktops
> bought on a bulk purchase deal.
>
Nothing about blade servers makes sense to me except that they are
cute and are more readily adapted for telcos.

I have yet to see the TCO argument about older PC's settled
definitively. Most calculations I've seen indicated, a few years ago
at least, that computers aged out of being cost effective for compute-
intensive operations after about three years. After that, you were
better off buying new than continuing to pay for electricity to run
obsolete equipment. Where the break-point is depends on the cost of
electricity, obviously. It also depends on the time derivatives of
watts/flop, of new-hardware-$/flop and of the cost of electricity.
From the POV of a departmental budget, though, maybe you don't care
about the cost of electricity.

Robert.