Hi Charlie, Always good to hear from you :-) On 19/05/09 06:50 PM, Charles E. Perkins wrote:
Hello folks, I've been following some of this discussion from afar, please excuse if I am not up to speed... Suresh Krishnan wrote:Hi Marcelo, On 17/05/09 08:58 AM, marcelo bagnulo braun wrote:ok, let me try again because it seems i am failing miserably to communicate what i want to say (or it simply so broken that people cannot believe what they are reading :-)I think it is a cool idea but it is very hard to realize in practice since it needs CNs to be changed.Right. This seems very much the point. If we are going to enable the use of flow labels, CNs need to be changed. The application that wants to have flows will have to run on the mobile and on the CN. But this seems perfectly fine, even desirable, to me. Am I missing something?
Yep. While I agree this is desirable and possible in the MNs, we are talking about the CNs.
For an application that wants to make a flow happen that doesn't use the flow label, is it somehow imagined that the application will not have to run on the CN?
This is not the point. If the application sets the flow label, the OS will honor it (at least in all the cases I have seen), but the point here is that the IPv6 stack needs to AUTOMATICALLY set the flow label for each new flow. i.e. every new connect() results in a new NON-ZERO flow label.
Suppose now that the app2 in the CN starts a flow to the MN. According to RFC3697, the CN SHOULD assign a different flow label to the packets belonging to this flow e.g. to this TCP connection.This is the crux of the issue. Almost every implementation I have seen sets the flow label to zero. This falls under the category "A source node which does not assign traffic to flows MUST set the Flow Label to zero.". Add to this the fact that compliance to RFC3697 is completely optional (it is not even mentioned in RFC4294), and this task quickly becomes pretty improbable.Implementation of what? Implementation of Mobile IP? Implementation of IPv6?
IPv6. Again, we are talking about CNs here. I do not see issues with the MN being updated to support this concept.
I guess the application would have to use something to indicate that the flow label should NOT be zero. I still am not really seeing what's wrong with that.
Nothing. Except that we are not talking about flow aware applications. We are talking about flow aware OSs automatically assigning flow labels.
In brief, since applications today do not use the flow label, I don't see what's wrong with the flow label being zero. When applications are built that need to use the flow label, they'll have to have operating systems support to do it.Thanks for any light you can shed on this.
Hope that clarifies my position. Thanks Suresh