#opendaylight-group-policy: group-policy-requirements
Meeting started by tbachman at 20:08:06 UTC
(full logs).
Meeting summary
- model presentation by Dvorkin (tbachman, 20:10:03)
- model is presented as a CLI-like concept
(tbachman,
20:10:23)
- model is less verbose, and aims for improved
clarity from yang model (tbachman,
20:10:45)
- contracts can be provided or consumed by
name (tbachman,
20:13:44)
- Additional “mutual” concept is for intra-EPG
contracts (tbachman,
20:14:30)
- groups can have requirements (optional),
capabilities (optional), and conditions (optional) (tbachman,
20:15:28)
- “mutual” is what we had previously called
“peer” contracts (tbachman,
20:15:59)
- contracts can also be provided/consumed using
expressions (tbachman,
20:16:20)
- expressions can use AND/OR/XOR/NOT-like
expressions (tbachman,
20:16:35)
- need to specify how the group relates to a
network domain. Can be subnet-domain, flood-domain,
bridge-domain (tbachman,
20:17:51)
- l3-subnet-domain has an address-range (e.g.
IPv4/v6 ranges), and can be part of a network domain or default L3
domain (tbachman,
20:19:10)
- contracts can be selected by one of two means:
by name, or by identifier (tbachman,
20:20:13)
- Walking through example (tbachman,
20:22:54)
- using tenant dave, with EPG deveness
(tbachman,
20:23:13)
- The default behavior is that two endpoints
within the same EPG can communicate w/o a contract (tbachman,
20:24:26)
- tenant dave also has another EPG
awesomeness (tbachman,
20:24:40)
- tenant dave also has a contract
pingability (tbachman,
20:25:02)
- The pingability contract has a subject, with a
classifier called reflexive icmp (tbachman,
20:25:27)
- clauses are implied in this case (tbachman,
20:26:05)
- mickey_spiegel says that you don’t have to
provide and consume to make it bi-directional (tbachman,
20:26:56)
- dvorkinista says that we can add a “peer”
contract to both provide and consume (tbachman,
20:28:51)
- mickey_spiegel says that classifier and action
have in/out/bi- in the UML model (tbachman,
20:29:16)
- to continue this example, something needs to
assign and endpoint to group daveness, and a second endpoint to
group awesomeness, and then the two endpoints can communicate with
the pingability (tbachman,
20:30:45)
- contract (tbachman,
20:30:51)
- lenrow asks what we mean by tenant (tbachman,
20:32:02)
- dvorkinista says that it can be thought of as
just a container (tbachman,
20:32:13)
- lenrow asks if it can have more than one L2
context (tbachman,
20:32:23)
- dvorkinista says yes (tbachman,
20:32:34)
- to extend the model, an l2-bridge-domain named
foo is added, which has part-of-domain mystuff (tbachman,
20:32:55)
- and an l3-domain mystuff is definied
(tbachman,
20:33:05)
- and the part-of-domain foo is added to the
daveness and awesomeness groups (tbachman,
20:33:48)
- The model is extended further by adding another
tenant (tbachman,
20:34:19)
- Another tenant called common is created
(tbachman,
20:34:41)
- The pingability contract is moved to the common
tenant (i.e. instead of being part of the dave tenant) (tbachman,
20:35:08)
- where the “common” tenant name is a special
name (tbachman,
20:35:20)
- This allows multiple tenants to share the
tenant common items (tbachman,
20:35:53)
- extending further, a new tenant named keith is
added (tbachman,
20:36:08)
- keith also has a group that references the
pingability contract defined in teh common tenant (tbachman,
20:36:35)
- question about AAA was raised (tbachman,
20:37:50)
- that needs to be defined in ODL first
(tbachman,
20:38:01)
- And then it can be used for GBP (tbachman,
20:38:18)
- mickey_spiegel asks if there are multiple
groups in scope, are we using longest-match prefix? (tbachman,
20:48:11)
- dvorkinista that can be one way (tbachman,
20:48:19)
- longest-prefix match (tbachman,
20:49:09)
- question asked for when policy is definied, is
that applicable to a specific topology that’s been reported?
(tbachman,
20:50:19)
- this is high-level intent — a policy
model (tbachman,
20:50:43)
- as an example, a high-level intent for network
virtualization (tbachman,
20:50:58)
- provides a mean of specifiying things like
access-control, without having to worry about what’s underneath,
which makes it more portable (tbachman,
20:51:59)
- The point is that the southbounds have
high-variability, so portability is important (tbachman,
20:52:25)
- question is asked what marries this policy to
low level things (tbachman,
20:53:23)
- the renderers provide this mapping (tbachman,
20:53:34)
- Question was asked if the peer contract
pingability was needed in teh amazingness and awesomeness groups,
given that both groups had part-of-domain greatstuff (tbachman,
20:58:43)
- the answer is that the l3-domain greatstuff is
addressability, but not reachability (tbachman,
20:59:02)
- so, the contract is needed for
reachability (tbachman,
20:59:09)
- what to do with the model presented by Dvorkin (tbachman, 21:03:23)
- Dvorkin just saw this as an instrument to
better explain how this works (tbachman,
21:03:39)
- mickey_spiegel notes that we can do the same
thing in yang that was done with this CLI-type model (tbachman,
21:05:35)
- readams notes that things like provide/consume
type relationships in yang is a bit tricky, as extraneous
information isn’t magically filtered out (tbachman,
21:08:43)
- mickey_spiegel asks if it makes sense to
demonstrate this with restconf (tbachman,
21:10:16)
- readams notes that you can do the same in
JSON (tbachman,
21:10:32)
- lenrow would like to see a user interface that
is easier to pick up and understand that what’s done in JSON
(tbachman,
21:22:55)
- readams notes that there would be some binding
on top of this to make it easier to use (e.g. python
bindings) (tbachman,
21:23:39)
- lenrow says that people care about the APIs,
not the bindings (tbachman,
21:23:54)
- readams notes that a new encoding could be
added to the northbound, if that’s found to be easier. (tbachman,
21:24:38)
- yang does have some constructs it can’t
support, which causes some of the constructs seen in the yang
model (tbachman,
21:34:24)
- ACTION: lenrow to
provide a link for a document of the UC&C use case (tbachman,
21:38:01)
- propose using Thurday status meeting for
continuation (tbachman,
21:39:45)
- update on steering/redirection (tbachman, 21:39:57)
- seeking more input from IETF SFC team
(tbachman,
21:40:15)
- need to propose way to integrate the rendering
needs of SFC into the GBP project (tbachman,
21:40:32)
- developing intro presentation for conversation
with IETF SF (tbachman,
21:40:55)
- SFC (tbachman,
21:40:59)
- UCIF requirements (tbachman,
21:43:45)
- UCIF requirements (tbachman, 21:43:47)
- UCIF QoS: A proposal for marking flows
(tbachman,
21:44:01)
- described in terms fo DSCP markings
(tbachman,
21:44:09)
- Need to propose intent expressions
instead (tbachman,
21:44:19)
- DSCP marking would be a rendering technique for
particular SB (tbachman,
21:44:35)
- http://www.ucif.org/Portals/0/documents/2014_02_27_Use_case.pdf
(tbachman,
21:45:06)
- UCIF Automated TE (tbachman,
21:45:13)
- Allows asynch updates to change marking for
specific flows (tbachman,
21:45:25)
- Allow explicit reallocation of bandwidth pool
between classes of service (tbachman,
21:45:42)
- Working on permission required to share in
ODL (tbachman,
21:45:48)
- discussion on need for realizing more use
cases (tbachman,
21:56:51)
- in order to uncover any possible corner cases
or issues (tbachman,
21:57:06)
- ACTION: lenrow to
propose continuing this discussion on Thursday’s status
meeting (tbachman,
22:00:24)
Meeting ended at 22:00:31 UTC
(full logs).
Action items
- lenrow to provide a link for a document of the UC&C use case
- lenrow to propose continuing this discussion on Thursday’s status meeting
People present (lines said)
- tbachman (93)
- odl_meetbot (4)
- alagalah (0)
Generated by MeetBot 0.1.4.