Re: A new project proposal vpp

Colin Dixon

So, it sounds to me like the project proposal wanted to try to factor out the common elements of how SFC, NetVirt, and GBP would talk to VPP to do service chaining int one place so that (a) the code could be shared and (b) we could reuse some of the work that GBP has done. However, the tone of this thread makes it pretty clear that (i) GBP isn't really interested in the shared code and (ii) is pretty opposed to the idea of their code being used as the base even if it's a separate copy.

Give that, it seems like the right path forward is for SFC and NetVirt to decide if they still want to have a shared code bas for how they interact with VPP and if that should be a new project.

I'll let that discussion keep going here and tentative schedule a creation review for next week—July 6th:


On Thu, Jun 22, 2017 at 8:02 PM, Yang, Yi Y <yi.y.yang@...> wrote:
Hi, Robert

vpp data store isn't mirror of data store in honeycomb/vpp node, they are some intermediate data, vpp renderer and classifier will renderer them to real vpp data in honeycomb/vpp node, I'll work out a PoC to show you what it looks like.

we hope GBP guys can agree this, if you check Srikanthi's PoC and RPCs GBP vpp renderer provided, you can assume the vpp data store is just to save those parameters for those RPCs, for classifier, NetVirt defined its own classifier, GBP defined its own classifier, SFC defined its own classifier, we hope they can have common classifier then augment it by itself, all the projects should use IETF ACLs and IETF interfaces, if so, the vpp data store can share these data in ODL.

-----Original Message-----
From: Robert Varga [mailto:nite@...]
Sent: Thursday, June 22, 2017 11:52 PM
To: Ed Warnicke <hagbard@...>; Yang, Yi Y <yi.y.yang@...>
Subject: Re: [Project-proposals] A new project proposal vpp

On 20/06/17 15:39, Ed Warnicke wrote:
> On Tue, Jun 20, 2017 at 4:25 AM, Yang, Yi Y <yi.y.yang@...
> <mailto:yi.y.yang@...>> wrote:
>     For vpp renderer, it needs to render bridge domain information
>     (neutron network, including VNI) and interfaces to bridge domain,
>     vxlan, vxlan-gpe, vhostuser etc in vpp node, you know information
>     model on ODL side is different from information model on VPP node
>     side, especially for sfc, vpp renderer will also render RSP to
>     bridge domain, vxlan-gpe, NSH entry and NSH map in every vpp node
>     RSP spans. sfc vpp renderer is much more complicated than NetVirt
>     needs, so we can use the existing code in sfc vpp renderer to
>     achieve a vpp renderer NetVirt needs very easily.
> I'm still not clear on what models are being rendered to what models.
> The SFC renderer renders the SFC models to the models being mounted
> via netconf.  It sounds like the *to* side of rendering is the models
> mounted via netconf.  What precisely are the *from* models?  Are we
> writing new models in this project for bridge domains, vxlan,
> vxlan-gpe, etc.  It also sounds like you are taking a small fragment
> of the SFC model and creating a new model for a bunch of virtual
> bridging concepts (similar to what is done in VBD).  Are there
> existing models being rendered to the netconf mounted models?

I tend to agree with Ed's point. I am probably missing something, as I am not too fluent in SFC/GBP information models nor how they align with NetVirt's view of the world.

I think we have multiple input models (SFC, netvirt, GBP) being rendered onto a target device (which happens to be VPP, but it's really about models).

What would be the models for "VPP data store"? Is this a mirror of the SB device's model and are we talking about the proposed project performing the equivalent of FRM?

I am sorry, but it is extremely hard to distill scope from the proposal in its current form. I think it needs to be beefed up with details of interactions and needs to be vetted by all of GBP, SFC and NetVirt to make sure the proposal actually fits the use case.

Furthermore it would be very beneficial to understand how this plays with netconf topology -- as I mentioned "VPP node" is in reality any SB device implementing a specific set of models (either directly or through an adaptor).


project-proposals mailing list

Join { to automatically receive all group messages.