#opendaylight-sfc: sfc_weekly
Meeting started by tbachman at 17:00:12 UTC
(full logs).
Meeting summary
- agenda (tbachman, 17:00:54)
- https://meetings.opendaylight.org/opendaylight-sfc/2015/sfc_weekly/opendaylight-sfc-sfc_weekly.2015-01-08-16.40.html
Last recorded SFC Meeting Minutes (tbachman,
17:01:08)
- presentation on scripts by Reinaldo (tbachman, 17:06:04)
- rpenno says that SFC gets its config from GUI
or REST to input configuration of service functions or service
function forwarders into the provider (tbachman,
17:07:48)
- the provider uses this information, along with
other sources, and creates an output of service function paths and
rendered service paths into the data store (tbachman,
17:08:17)
- the rendered service path describes the path
that will be taken when traversing the service functions
(tbachman,
17:08:36)
- The data store provides notifications to
listeners — e.g. openflow, REST, LISP (tbachman,
17:09:10)
- The notifications say that a new rendered
service path was created (tbachman,
17:09:32)
- This makes the architecture a point to
multipoint system, where the provider has no idea of who the
listeners are, and the listeners have no notion of who created the
information, making for a decoupled architecture (tbachman,
17:10:07)
- edwarnicke points out that each listener
figures out whether the notification applies to them (tbachman,
17:10:46)
- alagalah asks if everyone listens to the same
part, or do they listen to just their part, etc. (tbachman,
17:11:10)
- rpenno says that listeners subscribe to the
parts that their interested. They all should be interested in the
rendered service path (tbachman,
17:11:32)
- rpenno says there may be othe parts that
they’re interested in as well (tbachman,
17:11:41)
- ebrjohn says one thing that hasn’t been
addressed yet is that possible duplication could occur, depending on
what multiple listeners do (tbachman,
17:12:25)
- rpenno says the REST listener will use the
rendered service path and look at the switches it’s responsible for,
and will program all the switches (tbachman,
17:13:27)
- rpenno says the same things happen with deletes
and updates (tbachman,
17:13:59)
- listeners can save their own state in their own
data stores (tbachman,
17:14:11)
- rpenno says the demo he’s using has a rendered
service path with 3 switches — SFF1, SFF2, and SF3 (tbachman,
17:15:05)
- attached to each switch is a service function
(e.g. DPI, FW, NAT) (tbachman,
17:15:16)
- there are two service paths: ID1 and ID2
(tbachman,
17:15:29)
- These paths are symmetric (tbachman,
17:15:41)
- when a packet with ID1 is received, the SFF
sends the packet to SF1 (tbachman,
17:16:04)
- alagalah asks if there’s always a reverse
path (tbachman,
17:17:10)
- rpenno says that when you ask for a path to be
created, there’s a leaf node for indicating whether the path is to
be symmetric (true/false) (tbachman,
17:18:03)
- the reverse path would be the name of the
forward path plus “-reverse” (tbachman,
17:18:19)
- A service function forwarder has a data plane
locator and a dictionary (tbachman,
17:20:01)
- the dictionary indicates the services that it
can support (tbachman,
17:20:17)
- a service function chain does not tell you
where the packets are going to (tbachman,
17:20:34)
- the service funcion chain says packets will
traverse this chain, and the order of the services (e.g. FW, NAT,
etc.) (tbachman,
17:20:54)
- the service chain doesn’t indicate which
firewall it will traverse — only that it will traverse *a*
firewall (tbachman,
17:21:32)
- the service function path uses the “blueprint”
provided by a service function chain to create the actual path (e.g.
*this specific firewall*) (tbachman,
17:22:17)
- vina_ermagan asks when you compute the service
path, are they computed automatically using the existing service
functions (tbachman,
17:24:44)
- rpenno says yes — when you constuct a service
path, you look at the chain to see what is needed, and it goes
through all the candidate service functions, as well as which
functions are service by the service function forwarders with the
same transport, and if so, it uses those candidates to create the
rendered service path (tbachman,
17:25:44)
- demo (tbachman, 17:26:44)
- rpenno uses a python switch as the service
function listener for the demo (tbachman,
17:27:02)
- adds an SF and SFF to configuration, which
triggers notification to the python switch with NSH (tbachman,
17:27:20)
- rpenno says SFC has it’s persistence on it’s
own, independent of clustering, using config files (tbachman,
17:28:24)
- you first create the service function
forwarders, then configure the forwarders with their service
function dictionary (tbachman,
17:31:36)
Meeting ended at 17:55:37 UTC
(full logs).
Action items
- (none)
People present (lines said)
- tbachman (42)
- odl_meetbot (4)
- abhijitkumbhare (2)
- ebrjohn (2)
- alagalah (1)
Generated by MeetBot 0.1.4.