[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Egress Control (was Re: spc connections)

Zamfir,
        Please see below.

At 02:45 PM 11/18/2003, Anca Zamfir wrote: 
[...]
>There is no issue as long as you don't use different components/interfaces
>in each direction.

Agreed.

> If memory serves, this isn't supported in any case.

You mean not supported in the standard?

I am wrong. It's always bad to go from memory.

> I'm not sure what's missing, can you elaborate?

I was thinking that if LSP stitching is done at the egress node the two
directions could be on the same te link but different component links.
Egress node is a mid-LSR from the data plane perspective in this case. I
may be wrong so please correct.

I think this is a valid case and agree that specifying different components in each direction is not covered. There we could do this by using an unnumbered ero with a label in each direction. While this is possible, it wasn't what was originally envisioned.

>> > I think identifying components on transit links might be useful
>> > (particularly for RRO). But this may be implementation depended, i.e.,
>> > when a component link isn't also addressable as an unnumbered
>> > interface. Do you know of such a case?
>> >>when the Egress TE link is bundled. This is one of the
>> >>applications of ERO extensions specified in
>> >><<http://www.ietf.org/internet-drafts/draft-zamfir-explicit-resource-co
>> n
>> trol-bundle-02.txt><http://www.ietf.org/internet-drafts/draft-zamfir-explicit-resource-control-bundle-02.txt>http://www.ietf.org/internet-drafts/draft-zamfir-explicit-resource-control-bundle-02.txt><http://www.ietf.org/internet-drafts/draft-zamfir-explicit-resource-control-bundle-02.txt>http://www.ietf.org/internet-drafts/draft-zamfir-explicit-resource-control-bundle-02.txt.


>>
>>
>> >>
>> >>In this respect, do we agree that ERO extensions in the above mentioned
>> >>draft adds to the Egress control application?
>> >
>> >Again, not to egress control, but to explicit label/resource control on
>> >transit links.
>>
>>IMO, we should try if possible to have the same objects used for transit or
>>egress links.
>
>As I mentioned / asked before, is this a real case. Is there a case where
>component can't also be addressed as an unnumbered interface?

I don't know of a real case. A bundle by definition can have numbered,
unnumbered or a combination of component links. But I think that the
proposal you have works fine for all cases as long as same comp. link is
selected for both directions.

agreed. (I think!)

Thanks again for the comments.

Lou