[openstack-dev] [networking-sfc] What resources can and can't be reused

Duarte Cardoso, Igor igor.duarte.cardoso at intel.com
Mon Feb 13 21:26:54 UTC 2017


Hi Louis,

Yes, that makes sense - thanks for the feedback and the responses on my points.

Best regards,
Igor.

From: Henry Fourie [mailto:louis.fourie at huawei.com]
Sent: Monday, February 13, 2017 9:15 PM
To: OpenStack Development Mailing List (not for usage questions) <openstack-dev at lists.openstack.org>
Subject: Re: [openstack-dev] [networking-sfc] What resources can and can't be reused

Igor,
   For #6, the requirement on source-port for a flow-classifier is only for the OVS driver. This is not a restriction for other backend drivers.
In the case where there is no need for a sfc proxy to re-classify traffic returned from the egress port of a SF,
i.e., the SF is NSH-aware and it can receive, process and return the NSH, this restriction does not apply.
- Louis

From: Duarte Cardoso, Igor [mailto:igor.duarte.cardoso at intel.com]
Sent: Monday, February 13, 2017 12:27 PM
To: OpenStack Development Mailing List (not for usage questions)
Subject: Re: [openstack-dev] [networking-sfc] What resources can and can't be reused

Hi Cathy,

Relax only a couple of them. For Ocata I'm looking at disabling #6 if the chain/graph doesn't include sfc proxies (#6 seems to only be necessary if there are sfc proxies [1]). For Pike it would be interesting to make port-pair-groups completely reusable, as long as the flow classifiers don't make the choice of chain ambiguous.

[1] http://eavesdrop.openstack.org/meetings/service_chaining/2017/service_chaining.2017-01-12-17.14.log.html

Best regards,
Igor.

From: Cathy Zhang [mailto:Cathy.H.Zhang at huawei.com]
Sent: Monday, February 13, 2017 7:50 PM
To: OpenStack Development Mailing List (not for usage questions) <openstack-dev at lists.openstack.org<mailto:openstack-dev at lists.openstack.org>>
Subject: Re: [openstack-dev] [networking-sfc] What resources can and can't be reused

Hi Igor,

Before we dive into evaluation of the rules you listed below, I would like to understand whether you are suggesting to enforce the rules or relax the  rules/constraints you listed?
Could you clarify it?

Thanks,
Cathy

From: Duarte Cardoso, Igor [mailto:igor.duarte.cardoso at intel.com]
Sent: Monday, February 13, 2017 11:12 AM
To: OpenStack Development Mailing List (not for usage questions)
Subject: [openstack-dev] [networking-sfc] What resources can and can't be reused

Hi networking-sfc,

As part of my work regarding SFC Encapsulation and SFC Graphs, I exercised the API to understand exactly what resources can be reused, to possibly relax a few of the constraints when a chain is encapsulated end-to-end.
I'm requesting that the leaders and cores take a look at the list below, and reply if you see something that doesn't look quite right (or have any other comment/question). Thanks!

1. Every flow-classifier must have a logical source port.
2. The flow-classifier must be unique in its (full) definition.
3. A port-chain can have multiple flow-classifiers associated with exactly the same definition BUT different logical source ports.
4. The port-chains can be ambiguous, i.e. match on the same classification criteria, if and only if there are 0 flow classifiers associated.
5. The flow classifiers can only be used once, by a single port-chain.
6. Different port-chains cannot be associated to different flow classifiers that specify the same classification criteria BUT different logical source ports (this is https://bugs.launchpad.net/networking-sfc/+bug/1638421).
7. A port-pair's ingress cannot be in use by another port-pair's ingress.
8. A port-pair's egress cannot be in use by another port-pair's egress.
9. A port-pair can be associated to another port-pair's ingress and egress ports BUT swapped (i1=e2, e1=i2).
10. The port-pairs become "in use" when a port-pair-group associates them, so they can't be reused across port-pair-groups.
11. A port-chain can include port-pair-groups already associated to other port-chains, as long as not the exact same sequence as another port-chain (e.g. pc1: [ppg1,ppg2]; pc2: [ppg1] - is fine).

Best regards,
Igor.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20170213/ce80264e/attachment.html>


More information about the OpenStack-dev mailing list