Thanks Bob and Aaron for the replies. The problem with ToS is that it might be modified by routers in the path of the packet.
Collisions can also be problematic when an intermediate network uses tagged VLAN. Consider this, I have three OF networks connected via single links A---B----C. Also, all TCP connections are hop by. Now, my source is in A and destination is in C. The network B uses tagged VLAN while A and C does not. Here, if I use a VLAN tag to mark packets at network A, it is possible that network B will not forward them as intended (due to the VLAN tags). I am looking for a way to propagate the initial attachment point information <src ip, src port> across the network to the destination. One possible way is to attach an unique identifier to it and put that in a packet. The controller can maintain mapping between the identifier and the information. Is there another possible way to do this in OF? The ability to put an arbitrary identifier in a packet and to match on that identifier seems to be a good way to implement services on top of a network. Is there a reason this was not included in OF 1.0? Click allows the user to mark a packet using the Paint element and at the receiving end, the color can be checked with a CheckPaint element. Packets have a annotation field which can be accessed from userlevel. Paint sets that. Thanks On Fri, Nov 2, 2012 at 11:41 PM, Bob Lantz <[email protected]> wrote: > I think collisions are only problematic when you cannot disambiguate them. > > For example, packets coming from servers connected to different switch > ports can be disambiguated based on ingress port, irrespective of their IP > or Ethernet headers. > > Another example is NAT boxes - behind the NAT, everyone can be on a > different 10.x network reusing the same IP addresses. > > Whether you think of certain header bits as "color" or "smell" or "IP > address" or "forwarding hash" or "link-local virtual circuit identifier" is > basically up to you in an OpenFlow switch, except that you can do > longest-prefix matching on an IP address, and you may not be able to do so > on other fields like TCP ports. > > Does Click send color-tagged packets out switch ports in a way that can be > decoded at the other end of a link? If so, how are they encoded and decoded? > > On Nov 2, 2012, at 2:36 PM, Abhishek Chanda <[email protected]> > wrote: > > Hi all, > > Is there a way to insert custom metadata in a packet (other than the match > tuple) that can then be extracted in a switch? Specifically, I am looking > for something like Click's Paint element that allows marking packets with a > 'color'. Later, another network element can look at the color and use it to > decide forwarding behavior. > > Initially, I was thinking of using VLAN tags for this. But then, those > tags can collide with tags used in the underlying network and can create > confusion. > > http://www.read.cs.ucla.edu/click/elements/paint > > Thanks > _______________________________________________ > openflow-discuss mailing list > [email protected] > https://mailman.stanford.edu/mailman/listinfo/openflow-discuss > > >
_______________________________________________ openflow-discuss mailing list [email protected] https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
