#opendaylight-group-policy: gbp_arch
Meeting started by tbachman at 17:01:35 UTC
(full logs).
Meeting summary
-
- dlenrow thinks of virtual functions as EPGs,
which can have contracts with other EPGs (tbachman,
17:13:36)
- an example contract can be send IP traffic to
the internet, and service chain EPG would provide a contract to
handle the chain to send to the internet (tbachman,
17:14:40)
- dlenrow asks whether EPGs can support
sEPGs (tbachman,
17:14:58)
- recursion in the model is modeled as a linked
list (tbachman,
17:15:10)
- due to a yang limitation (tbachman,
17:15:16)
- dlenrow thinks of chains as an ordered list of
EPGs (tbachman,
17:15:39)
- dvorkinista says that the goal here is to
provide intent, (tbachman,
17:16:08)
- and he thinks that enforcing the EPG in the
hypervisor might not be the right thing to do. (tbachman,
17:16:22)
- b/c you might not be able to satisfy the
constraints in the hypervisor (tbachman,
17:16:36)
- it would be nicer if we can think of a service
chain as an ordered graph to which traffic is subjected (tbachman,
17:17:06)
- like filters or transformers (tbachman,
17:17:13)
- dvorkinista says these are functions you
subject traffic to (tbachman,
17:17:30)
- (i.e. apply subject or set of subjects to the
conversation) (tbachman,
17:17:53)
- dlenrow says that the job of assigning which EP
in an EPG to send the traffic through is a job of the
renderer (tbachman,
17:19:14)
- dvorkinista says this is essentially correct,
but w/o a network address involved (tbachman,
17:19:44)
- paulq says that as long as you have a locator,
that’s enough, regardless of locator (tbachman,
17:20:16)
- dvorkinista says that locator is an
implementation detail. (tbachman,
17:20:25)
- as an example, a firewall followed by a load
balancer, he doesn’t care what device, IP address, etc. (tbachman,
17:20:56)
- b/c the goal is to capture the intent in a
portable way (tbachman,
17:21:08)
- dlenrow says that there’s still a policy that
says that you want to go through something (e.g. firewall)
(tbachman,
17:24:26)
- dvorkinista says it’s not a EP
abstraction (tbachman,
17:24:39)
- where the abstraction is a service graph of
logical functions, which can map 1:1 to a box, or N:N (tbachman,
17:24:57)
- dlenrow asks if we agree that from the
network’s perspective, a function is something that we have to send
packets throuugh (tbachman,
17:25:31)
- dvorkinista says not necesarrily (tbachman,
17:25:38)
- if it’s done in a hypervisor, you don’t
redirect anything (tbachman,
17:25:49)
- b/c it’s all confined to the virtual
switch (tbachman,
17:25:56)
- dlenrow says that the virtual function still
has an IP address (tbachman,
17:26:10)
- dvorkinista says this can be enforced right in
the hypervisor, like nicria/NSX zone-based security (tbachman,
17:26:42)
- dlenrow says that in an SDN domain, this is all
enforced by a controller (tbachman,
17:27:05)
- https://cisco.webex.com/mw0401l/mywebex/default.do?siteurl=cisco&service=1&main_url=%2Fmc0901l%2Fmeetingcenter%2Fdefault.do%3Fsiteurl%3Dcisco%26main_url%3D%252Fmc0901l%252Fmeetingcenter%252Fmeetingend%252Flandingpage.do%253Fsiteurl%253Dcisco%2526ishost%253Dtrue%2526NM%253Dkrb%2526AD%253Dkrb%2540cisco.com%2526STD%253D1&rnd=-2133996819
(alagalah,
17:29:09)
- ACTION: dvorkinista
will write up a sketch of the model he’s thinking, which can be
discussed on monday or some other meeting (tbachman,
17:29:14)
- dlenrow says This is the nature of GBP
hierarchical model. Operator admin defines some VFs and EPGs. tenant
user is aware of EPGs like internet and engineering. Operator guy is
aware of EPGs that are VFs. One simple uniform concept for all
reqardless of sophistication (tbachman,
17:37:57)
Meeting ended at 17:41:02 UTC
(full logs).
Action items
- dvorkinista will write up a sketch of the model he’s thinking, which can be discussed on monday or some other meeting
People present (lines said)
- tbachman (57)
- dlenrow (14)
- paulq (4)
- mickey_spiegel (4)
- odl_meetbot (3)
- dconde (2)
- s3wong (2)
- hemanthravi (2)
- alagalah (1)
Generated by MeetBot 0.1.4.