right?
Yes, that's right.
Also, when I say 'same LLDP packet', I am thinking that SW2 should
forward the LLDP packet of SW1 as received from SW1 and not of itself
i.e SW2.
Yes.
--
Wes Felter
IBM Research - Austin
___
openflow-discu
Once LLDP packet is on wire,is the rule [2] intrinsically defaulted
in the switch?
Generally yes. It depends on what version of OpenFlow you're using.
In general, a controller does not need to install any flows to perform
LLDP discovery. The controller can just use packet out and packet in
On 3/11/14 12:50 PM, Slavica Tomovic wrote:
I am using Reference OpenFlow software switches in my testbed.
Try OVS 2.0. Also try using a faster processor, like a Celeron.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
Yes, the connection is bidirectional.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
See
https://wiki.opendaylight.org/view/OpenDaylight_Controller:Java_API_Reference
but other controllers might have better API documentation.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
y be able to do this with meters.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
in this, please?
What part do you need help with? Do you understand how IP source guard
works? Do you have Mininet and POX installed?
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https
nd of QoS that meters and queues cannot implement?
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
some of the network switches and a
other controller responsible for the remaining.
In theory OF switches should be managed using the OF-Config protocol and
someone should write an OF Manager, but I don't think anybody is
actually working on it. You could write one and become a hero... :-
.
In an L3 network the number of flows might be equal to the number of
subnets, so a small number of updates per second might be acceptable.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https
rate. The observed flow setup rate will
depend on what the controller is doing; for example a network with
proactive destination-based forwarding would have almost no flow setups.
Until a switch fails and there will be a burst of 1,000 flow mods per
second.
--
Wes Felter
IBM Researc
as well as layer 2 switching.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
flow mods, not group table entries.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
I think, standard MSTP can be integrated with OpenFlow controller with
packet-in and packet-out as mentioned by Wes Felter with some
modification in flood mechanism (per Vlan port state lookup before
flood) as explained in my first post but it will not get OF spec support
as basic 802.1D STP
ntrol plane and entirely depend
on controller.
This is already possible using packet in and packet out.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman.stanford.edu/mailman/lis
allows routing between them)?
The networks can speak existing routing protocols (OSPF, IS-IS, BGP,
etc.) to each other. The controllers would implement the routing control
plane. See RouteFlow as an example.
--
Wes Felter
IBM Research - Austin
/ cons of such an approach.
It could be more reliable. It could be faster (due to having more
hardware) or it could be slower (due to running Paxos all the time).
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss
you can't match on it. Maybe you should describe the big picture of what
you're trying to accomplish and we can give you some advice.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@list
Ideas or suggestions are welcome.
There have been about 50 papers on this topic at SIGCOMM, NSDI, HotNets,
HotSDN, CoNext, ANCS, etc. in the last three or four years. If you read
the latest papers you'll probably find plenty of ideas for improvements.
--
Wes Felter
IBM Research - A
odules.
http://docs.projectfloodlight.org/display/floodlightcontroller/Configuration+HOWTO
You can probably get better help with Floodlight from the Floodlight
mailing list because there are more people there.
--
Wes Felter
IBM Research - Austin
___
openflow-di
ports.
OpenFlow also supports PBB but I don't know if it's implemented anywhere.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
ucket in the group, based on a switch-computed selection
algorithm (e.g. hash on some user-configured tuple orsimple round
robin).
Yes, select groups should be implemented with ECMP logic (if any switch
ever supports them). This will probably be clearer once the L2/L3 TTP
gets released.
--
Wes F
n" routing protocol based on Paxos.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
igTap both
implement this kind of feature.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
flow when i need.
Right. The controller should not close an OpenFlow connection because
the switch will just reopen it again.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman
low and a track
record of publishing their research in top tier research venues.
Please send resumes to me immediately.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman.stanford.edu
ow Mods/Second
1 722
5 1219
10 1392
20 1568
Another figure of merit is the number of packet ins and outs per second
the switch can handle; the G8264 can generate about 200 packet ins/s.
--
Wes Felter
IBM R
ble demand (which depends on control granularity). We
discuss this topic in our CoNEXT paper where we use exact-match
destination-based forwarding to reduce table demand.
http://conferences.sigcomm.org/co-next/2012/eproceedings/conext/p49.pdf
--
Wes Felter
IBM Researc
ogy so it isn't
predefined. Once the topology is known, routing can be done with
Dijkstra's algorithm.
Traditional routing protocols like OSPF or BGP are generally not needed
inside an OpenFlow domain, but if you want to speak those protocols to
the outside world, look at Route
read that
code to see how it works.
--
Wes Felter
IBM Research - Austin
___
openflow-discuss mailing list
openflow-discuss@lists.stanford.edu
https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
not have an ARP table. Your controller needs
to snoop all ARPs and track host locations itself. Once you've done
that, it sounds like you know how to build the appropriate forwarding
tree(s).
--
Wes Felter
IBM Research - Austin
___
openflow-disc
true, how about the ACKs used in TCP layer?
A TCP ACK only indicates that the packet was received by the TCP stack;
it does not indicate that any processing has been done. Also, TCP ACKs
aren't really visible at the socket layer, so the controller won't even
know if the data has been AC
nning of OpenFlow time:
http://www.openflow.org/documents/OpenFlow_1_1_Multiple_Tables_06_22_2010.pdf
Also, switch chips have multiple tables, so it seems simpler to expose
them as multiple tables. Since no one has bothered to implement this we
haven't yet discovered how painful it will be
.
If the answer is "yes", then what if applications require different MAC
or even PHY mechanisms?
OpenFlow 1.x is somewhat biased towards Ethernet; people should make
proposals to ONF if they want support for other protocols. Or maybe you
can just use OXMs.
35 matches
Mail list logo
