Re: hwvtep high level control flow

daya k

hi anil,
sounds good. i will add it in the workflow and send out before the meeting.

for #1, i can add a workflow for hwvtep-hwvtep tunnels, however, it does not really fit in with the gateway model. but technically, should be supported.
for #2, i have hinted at this scenario where switch initiates the connection on slide 1 bullet 2. basically, if the switch has already actively connected to the controller, the controller needs to co-relate an l2-gw create operation with an existing node. in this case, we need to use some mechanism to co-relate the device id received from the api with a node, so my question to the community would be - can we ensure device_id is set to switch name in the NB, and we use this to co-relate? 
ravi, is there some other mechanism used in the openstack version of l2gw today?


From: Anil Vishnoi <vishnoianil@...>
To: Ravi Sabapathy <Ravi_Sabapathy@...>
Cc: "daya_k@..." <daya_k@...>; Sam Hague <shague86@...>; Andre Fredette <afredette@...>; "Kenchappa, Ravindra" <ravindra.kenchappa@...>; "ovsdb-dev@..." <ovsdb-dev@...>
Sent: Monday, August 31, 2015 3:40 AM
Subject: Re: [ovsdb-dev] hwvtep high level control flow


It's good to add md-sal data store in between, because in beryllium release neutron-northbound project won't be talking to net-virt directly. Neutron northbound will be pushing all the data to data store and net-virt will be listening to those data changes.


As of now Net-Virt relies that use will connect the compute nodes to the controller using tcp:ip:port, rather then putting switch in passive mode (ptcp:port) and expected controller to connect to it. Although OVSDB southbound plugin allows user to proactively connect to the switch if switch is listening for the connection by putting hyper-visor data in config data store. If OpenStack l2gateway rest APIs provide hardware vtep device information (ip,port this device is listening for connection on) then i think you can support controller initiated connection as well.


On Fri, Aug 28, 2015 at 8:36 AM, <Ravi_Sabapathy@...> wrote:
Hi Daya/Sam,
              I have the few queries on the document.
1. There are 3 ways VXLAN can be established,
i.         VXLAN between OVS switches controlled by openstack and ODL.
a.       This is already supported in software VTEP.
ii.       VXLAN between hardware gateway and OVS switch controlled by openstack and ODL.
a.       This will be supported in hardware VTEP project.
iii.     VXLAN between 2 hardware gateways connected to bare metal servers.
a.       Are we planning to support VXLAN between hardware gateway. From the control flow document, I could infer we don’t support this kind of a setup.
        2. How does ODL react when the switch initiates the OVSDB  connection using ptcp?
I am asking this because the switches will have CLI’s to do these configurations and it is possible the switch can initiate the OVSDB connection. As per the L2GW slide, the openstack initiates the hardware gateway creation.
If we are planning to support this case, will the openstack be informed about the hardware gateway creation?
From: ovsdb-dev-bounces@... [mailto:ovsdb-dev-bounces@...] On Behalf Of daya kamath
Sent: Thursday, August 27, 2015 5:39 PM
To: Sam Hague <shague86@...>; Anil Vishnoi <vishnoianil@...>; Andre Fredette <afredette@...>; Ravindra Kenchappa <ravindra.kenchappa@...>; Ovsdb-dev <ovsdb-dev@...>
Subject: [ovsdb-dev] hwvtep high level control flow
here's a first draft of how the control flow to incorporate hwvtep into netvirt.
sam, unfortunately, out of force of habit i started in ppt, and when i copied it to google, it got pasted as a picture.
attaching to this email also, i can incorporate comments and update, or re-do it so its editable, when i have some more time.


Join { to automatically receive all group messages.