#opendaylight-group-policy Meeting
Meeting started by regXboi at 18:02:46 UTC
(full logs).
Meeting summary
- architecture meeting (regXboi, 18:02:56)
- mickey_spiegel asks that since architecture
touches the MD-SAL a lot, are people with comfortable with using the
MD-SAL (regXboi,
18:04:33)
- For now, assume MD-SAL for purposes of
discussion. Data store subgroup will figure out if we actually go
with MD-SAL (mickey_spiegel,
18:04:59)
- jmedved said we need to find appropriate
sharding mechanisms (dconde,
18:08:20)
- regxboi says it makes sense and he spoke w/
Colin (dconde,
18:08:54)
- jmedved wants to leverage a lot of the work the
APIC team has done, seems like the idea is AKKA would create the
shards and leverage APIC's model of consistency (alagalah,
18:09:13)
- dvorkinsta state we need to figure out how the
requirements we have on MD-SAL and whether it satisfied our
requirements. (dconde,
18:10:08)
- we need to do that quickly. (dconde,
18:10:16)
- WE need to understand if AD-SAL can provide
that. (dconde,
18:10:51)
- edit: above should be MD-SAL (regXboi,
18:11:19)
- jmedved says he has a interesting architectural
blueprint for the MDSAL (alagalah,
18:11:23)
- concerning using akka for knitting together
shards that are stored within existing in-memory structures
(regXboi,
18:12:57)
- sharding via Akka. Matthew will have
preentation on Tuesday. need experimentation to see (dconde,
18:12:59)
- want to take a step back (dconde,
18:13:13)
- ask whether to go w/ MD-SAL or not.
(dconde,
18:13:27)
- if not, what do we use ? (dconde,
18:13:45)
- the alternative, regxboi states, is that we
better come up with alternative. (dconde,
18:14:04)
- SAL is the backbone of the controller.
(dconde,
18:14:23)
- so if we do not use it (model driven or not),
then we are building a different system. (dconde,
18:14:40)
- the msg bus lies within SAL. (dconde,
18:14:52)
- note: above three statements are from
jmedved (regXboi,
18:15:11)
- the APIs are eventually provided in the SAL. IF
not, we need to build all these mechanisms ourselves. (dconde,
18:15:21)
- regxboi says let's make explicit. DATASTORE
group was asked with a way to store policies in a datastore
(dconde,
18:16:08)
- that is different from whether MD-SAL is the
routing backbone. (dconde,
18:16:18)
- if we need to the DATASTORE people to take that
task, it's a separate issue. (dconde,
18:16:45)
- jmedved says SAL can be made independent of
data store, and stll be within architectural footprint of controller
and (dconde,
18:17:24)
- we lost Jan on audio (dconde,
18:18:16)
- regxboi says rate and amount are
important. (dconde,
18:19:15)
- dvokinista states JMedVed's Mac crapped
out. (dconde,
18:19:38)
- mickey_spiegel states thhat datastore's
requirements are more important than SAL. (dconde,
18:20:34)
- not more important, I just thought that comes
first, whereas Jan is looking at it the other way around
(mickey_spiegel,
18:20:56)
- dvorkinista says let's don't do it bottoms
up. (dconde,
18:20:59)
- alagalah says that it is in Hydrogen
(dconde,
18:21:32)
- regxboi say we hope that it gets better in
Helium (perf wise) (dconde,
18:21:51)
- Jmedved is back. (dconde,
18:21:58)
- jmedved MD-SAL perf #s. (dconde,
18:22:20)
- one is pure RPC notifivation perf
CBENCH. (dconde,
18:22:39)
- readams asks where we want to use these
components. (dconde,
18:24:07)
- dvorkista say as long as it's fast enough it's
good t (dconde,
18:24:37)
- quick 101 of MD-SAL (regXboi,
18:25:22)
- one piece jmedved says is missing in MD-SAL is
ability to do queries (regXboi,
18:27:48)
- dvorkinista wants to use policy requirements to
drive how MD-SAL does queries (regXboi,
18:28:28)
- jmedved agrees (regXboi,
18:28:33)
- jmedved claims on RPC that performance is
similair to floodlight (regXboi,
18:29:13)
- edit: similar (regXboi,
18:29:24)
- mickey_spiegel asks about details (regXboi,
18:30:10)
- about subscribing to subtrees (mickey_spiegel,
18:30:24)
- jmedved says we can, but he's not prepared to
do it right now (regXboi,
18:30:24)
- mickey_spiegel asks about classes versus
objects (regXboi,
18:30:52)
- jmedved says the yang model defines the api
that are generated (regXboi,
18:31:08)
- but he's not sure whether subscription is to
class or object (regXboi,
18:31:59)
- sounds like object is currently supported,
class is not yet supported, requires some development that can be
done (mickey_spiegel,
18:33:20)
- Ryan is OK with the idea of having an
implementation of the data store that is packaged with the
controller and ships (mickey_spiegel,
18:36:57)
- Jan notes already have two implementations of
data store (mickey_spiegel,
18:37:23)
- mickey_spiegel asks about NB APIs (regXboi,
18:39:35)
- Jan answers that you define the NB API in yang
and it gets translated into RESTCONF (regXboi,
18:40:07)
- dvorkinista asks for clarity that every object
gets an API (regXboi,
18:41:18)
- jmedved says yes and that is a plus and a
minus (regXboi,
18:41:31)
- because there aren't access policies
today (regXboi,
18:41:46)
- discussion that access control will be applied
to trees (regXboi,
18:42:50)
- today everything is open (regXboi,
18:42:57)
- no RBAC today... that needs to be done
(regXboi,
18:43:32)
- Need discussion with OpenStack group-based
policy at some point to make sure ODL RESTCONF is compatible
(mickey_spiegel,
18:47:53)
- also need to nail down the NB API yang early in
the game to launch that discussion (regXboi,
18:48:21)
- mickey_spiegel question is how to bind to
neutron (dconde,
18:52:00)
- port bindings in a normal way vs. GBP?
(dconde,
18:52:18)
- s3wong say absolutely we want ot use ODL GBP
without neutron GBP. (dconde,
18:52:43)
- dvorkistia requests agenda for Friday
meetings. (dconde,
18:53:35)
Meeting ended at 18:55:20 UTC
(full logs).
Action items
- (none)
People present (lines said)
- dconde (39)
- regXboi (33)
- mickey_spiegel (9)
- odl_meetbot (5)
- alagalah (4)
- s3wong (2)
- mestery (1)
- readams (1)
Generated by MeetBot 0.1.4.