[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [nvrg-bof] Updated charter proposal



-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1



James Kempf wrote:
> Hi Joe,
> 
> Thanx for the reply. Regarding: 
> 
>>> Currently,
>>> there is much work going on in GENI on control frameworks for 
>>> provisioning a virtual aggregate (a network slice and compute 
>>> resources).
>> This is not network virtualization. I.e., when we deal with 
>> an OS, we don't try to standardize the provisioning of 
>> memory, processing capacity, etc. Processes don't ask for 20% 
>> of a CPU; they just ask to run. Processes don't ask for 20% 
>> of memory, they ask for it in terms they need (bytes). 
>> Similarly, virtual networks ask for resources in terms they 
>> need (addresses, links, and reserved link capacity, i.e., 
>> provisioning). We need a common way to express how to 
>> virtualize the network (which most "virtualization" systems 
>> don't really do anyway) before we can standardize how its 
>> resources are manged.
> 
> GENI has the notion of RSPECs which I believe are designed to allow
> virtual network providers to ask for resources. There has also been some
> recent work in 4WARD along these lines. Are you saying that the way
> these proposals are designed doesn't really address the need or perhaps
> that they need some modification?

RSPECs, like many things in GENI, focus on services and computation -
i.e., OS resources, not network resources (e.g., addresses, interfaces,
virtual links, etc.) - at least according to the March 2008 slides I
could find on the topic.

> One of the issues we've been discussing here at ER is that there is
> already a collection of deployed data plane techniques to support
> network virtualization. For dedicated bandwith, the techniques run from
> dedicated lambdas up to MPLS LSPs in IP networks and PBB-TE tunnels in
> Metro Ethernet. For nondedicated bandwidth, there are IP-IP tunnels,
> maybe one could even view HIP as a kind of tunnel. Are these techniques
> insufficient?

There are lots of building blocks, and a few systems that create network
architectures out of them (which is more than just the building blocks).
We need to develop a common framework for virtualizing the *network*;
then we can engage groups like GENI about how to organize distributed
resources on it.

> If you grant that these techniques are sufficient, then the primary
> issue becomes how to  construct a virtual network in a cross provider
> way. This today, as far as I can determine, is impossible except if you
> run over IP. 

It can be done at IP, MPLS, or ethernet, as well as some other less
ubiquitous ways today.

> The issue you cite above, namely how to name and allocate
> resources, is a big one, also interoperability between providers to
> allow the slices to be plumbed together.

Slices, AFAICT, refer to a group of (virtualized) OS resources. It's not
useful to talk about those resources until the *network* resources are
virtualized and coordinated.

> There is also a question of
> timescale. Most providers require a timescale of days to set up a
> virtual network like a VPN. I believe the vision of most researchers is
> such virtual networks to be available on demand.

You're jumping a few steps ahead. IMO, the steps are:

	- define a common virtual network framework
	(what this BOF is for)
	- determine how providers can offer VNFs as a service
	- determine how to provision a VNF across providers
	- integrate VNFs with slice reservation systems

We're at step one here.

> So our view on this is that the work is primarily needed in constructing
> a control plane that allows virtual network slices to be constructed
> from resources obtained from different ISPs by a virtual network
> provider. And then to connect the network resources up to end host
> resources (which may involve clouds or end host virtual machines). This
> seems to me what GENI is trying to address, maybe Trilogy and Akari as
> well (I am not that familiar with them).

It's hard to consider discussing a control plane when we haven't yet
defined what to control. Slices don't include network interfaces or
addresses as a virtualized, localized resources - that is the purpose of
network virtualization, and it is useful independently of a slice. So
let's start by decoupling the two, and solve the network virtualization
problem first.

Joe
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkowBQgACgkQE5f5cImnZru+lACg3yyM6NNqYz90irCUtrFIZF+9
QvAAn0zElAT+gaKjYR1dRruVq4Z68Xhr
=0bYe
-----END PGP SIGNATURE-----


Note Well: Messages sent to this mailing list are the opinions of the senders and do not imply endorsement by the IETF.