[vnrg] Layering vs. virtualization

Roland Bless <roland.bless@kit.edu> Thu, 29 July 2010 08:20 UTC

Message-ID: <4C5139C6.9020300@kit.edu>
Date: Thu, 29 Jul 2010 10:20:22 +0200
From: Roland Bless <roland.bless@kit.edu>
Organization: Institute of Telematics, Karlsruhe Institute of Technology (KIT)
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.8.0.1) Gecko/20060111 Thunderbird/1.5 Mnenhy/0.7.3.0
MIME-Version: 1.0
To: vnrg@irtf.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
Subject: [vnrg] Layering vs. virtualization
Precedence: list

Hi,

I thought a little bit more on that topic, e.g., whether IP
on top of Ethernet itself is a virtualization technique and I
think that it is not - longer rationale below. First, I want to second
what Aaron said: we should consider network technologies other
than IP in the substrate as well as in the virtual network. So
sometimes it is easier to think of some abstract substrate technology
instead of IP as substrate.

However, I think one difference between "layering" and virtualization
is as follows:
when you put another protocol layer on top, you usually have to
do it in the "end-system"/at the end points, i.e., IP nodes are
sitting at L2 end points whereas L2 nodes (e.g., switches) are
transparent for L2 end points.
Same in L3 (letting the control plane aside for this moment):
routers as L3 network nodes are largely transparent for
the end-systems (except for first/last-hop routers),
i.e., a transport connection at L4 is normally
terminated in L3 end-systems. So in this way neither IP is a virtual
network on top of Ethernet nor is a TCP connection on top of IP, but I
would consider IP as an overlay and abstraction technique (it mainly
abstracts from different L2 networks in its substrate).

In contrast, a virtualization technique in/at L2 involves mechanisms
within the L2 nodes, e.g., support of VLAN tagging.
A real network virtualization technique at layer 3 would require,
e.g., partitioning of a L3 node/a router; lets consider that you are
running a different protocol than IP in a partition.
The "hard part" now is getting/demultiplexing from the substrate
to the virtual parts of the router. There are various ways to do it
depending on the substrate's capabilities.
So using a dedicated physical L2 port would be one possibility,
using VLANs over a shared L2 cable would be another. If the substrate
is on higher layers MPLS LSPs or L3 tunnels etc. can be used.

Sometimes it also helps to think on addressing the virtual resources
from the control plane inside the substrate. Basically you have
to address a VNet (denoting a specific virtual network),
a Virtual Node, and a specific Virtual Interface inside the
Virtual Node, e.g., in order to connect a substrate link/tunnel
to a specific interface of this particular virtual node.
However, it is not required that VNet-IDs, Virtual Node IDs,
or Virtual Link/Interface IDs are literally carried in substrate
data packets since there could be link-specific mapping techniques using
available multiplexing mechanisms, e.g., VLAN-tags.
In analogy one can denote such link-specific identifiers for VNets as
"VNet-Tags". A VNet-Tag identifies a virtual link in a link-specific
context. In absence of multiplexing support in the substrate, it may be
required to use an explicit shim header that carries the
VNet-Tag in order to allow proper demultiplexing of virtual networks on
a shared substrate link.

To keep a long story short: when talking about virtualization
we must be specific which layer is actually virtualized or do we
consider layer 3 only?

Regards,
 Roland

[vnrg] Layering vs. virtualization Roland Bless
Re: [vnrg] Layering vs. virtualization Pedro Andrés Aranda Gutiérrez
Re: [vnrg] Layering vs. virtualization Monique Morrow (mmorrow)
Re: [vnrg] Layering vs. virtualization Roland Bless
Re: [vnrg] Layering vs. virtualization Pedro Andrés Aranda Gutiérrez