[ovs-discuss] [openstack-dev] [OVN] [networking-ovn] [networking-sfc] SFC andOVN

Ryan Moats rmoats at us.ibm.com
Fri Jun 17 14:26:46 UTC 2016


Apologies for being delayed on replying and in-line back as well

Ryan

John McDowall <jmcdowall at paloaltonetworks.com> wrote on 06/15/2016 05:58:35
PM:

> From: John McDowall <jmcdowall at paloaltonetworks.com>
> To: Ryan Moats/Omaha/IBM at IBMUS
> Cc: Na Zhu <nazhu at cn.ibm.com>, Srilatha Tangirala/San Francisco/
> IBM at IBMUS, "OpenStack Development Mailing List (not for usage
> questions)" <openstack-dev at lists.openstack.org>, discuss
> <discuss at openvswitch.org>
> Date: 06/15/2016 05:58 PM
> Subject: Re: [ovs-discuss] [openstack-dev] [OVN] [networking-ovn]
> [networking-sfc] SFC andOVN
>
> Ryan,
>
> In-line
>
> Regards
>
> John
>
> From: Ryan Moats <rmoats at us.ibm.com>
> Date: Tuesday, June 14, 2016 at 9:42 PM
> To: John McDowall <jmcdowall at paloaltonetworks.com>
> Cc: Na Zhu <nazhu at cn.ibm.com>, Srilatha Tangirala <srilatta at us.ibm.com
> >, "OpenStack Development Mailing List (not for usage questions)" <
> openstack-dev at lists.openstack.org>, discuss <discuss at openvswitch.org>
> Subject: Re: [ovs-discuss] [openstack-dev] [OVN] [networking-ovn]
> [networking-sfc] SFC andOVN
>
> "discuss" <discuss-bounces at openvswitch.org> wrote on 06/14/2016 10:31:40
PM:
>
> > From: John McDowall <jmcdowall at paloaltonetworks.com>
> > To: Na Zhu <nazhu at cn.ibm.com>
> > Cc: Srilatha Tangirala/San Francisco/IBM at IBMUS, "OpenStack
> > Development Mailing List \(not for usage questions\)" <openstack-
> > dev at lists.openstack.org>, discuss <discuss at openvswitch.org>
> > Date: 06/14/2016 10:48 PM
> > Subject: Re: [ovs-discuss] [openstack-dev] [OVN] [networking-ovn]
> > [networking-sfc] SFC andOVN
> > Sent by: "discuss" <discuss-bounces at openvswitch.org>
> >
> > Juno,
> >
> > It is a container for port-pair-groups and flow-classifier. I
> > imagine there could be more the than one port-chain per switch. Also
> > we may want to extend the model beyond a single lswitch
>
> I agree that there could be more than one port-chain per switch,
determined
> by the flow classifier.
>
> What I'm confused by is:
>
> 1. Why are items only recorded in logical switches?  I would think
> that I could also attach an SFC to a logical router - although I admit
> that the current neutron model for ports doesn't really allow that
> easily.  Couple that with the change of name from Logical_Port to
> Logical_Switch_Port, and I'm left wondering if we aren't better off
> with the following "weak" links instead:
> -the Port_Chain includes the logical switch as an external_id
> -each Port_Pair_Group includes the Port_Chain as an external_id
> -each Port_Pair includes the PPG as an external_id
> -each Logical_Switch_Port includes the PP as an external_id
>
> I *think* that *might* allow me (in the future) to attach a port chain
> to a logical router by setting the logical router as an external_id and
> using Logical_Router_Ports to make up the PPs...
>
> JED> If there are “port-chain” tables for switches and routers I
> think I agree. Not sure how this is impacted by the type of VNF (see
> the last email to Juno). I struggle a bit with imagining the flows.

RM> Back in the day when we discussed this internally here, SFCs could
RM> be inserted as BiW (which we do a good job with currently) and at
RM> network boundaries (which I'm not sure how I could do with the
RM> current model) - my router question is more one of leaving the
RM> door open for the boundary case (sorry for the pun) in the future.

> 2. I still don't see what Logical_Flow_Classifier is buying me that
> ACL doesn't - I can codify all of the classifiers given in the match
> criteria of an ACL entry and codify the first PPG of the SFC as
> the action of the ACL entry...
> JED> Flow classifiers do map to an ACL entry – just need additional
> metadata, I.e. Action of the ACL and wether the rules should be uni
> or bi-directional. Though that information could be in the port-chain.

RM> yes and I see the action field of the ACL table being extended
RM> to include "enter port chain <blah>" to cover that metadata.
RM> Why couldn't bidirectional Flow Classifiers at SFC just be
RM> implemented as a pair of uni-directional ACLs in the NB DB?
RM> I'll back off on this point if I can see an example of an flow
RM> classifier that can't be made to fit in the ACL table, but to
RM> date, I've not been able to construct such a beast.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20160617/d0f09aee/attachment-0002.html>


More information about the discuss mailing list