The MPLS WG Archive

[Roch Guerin <guerin@watson.ibm.com>]

>could you elaborate what the problem is?

Juha,

Here is a possible scenario along the lines of what I had in mind:

    1
S1-----\                             You have 4 senders, S1 to S4, headed to
        \___                         a common egress point as indicated on the
    1   /M1 \                        diagram on the left.  They each want one
S2-----/     \2  4 (3)               unit of bandwidth, so that after each merge 
              \_______   Common link point the requests are added, i.e., 2 units
    1         /M3        to egress   on the links after merge points M1 and M2, 
S3-----\     /2                      and 4 units after merge point M3.  However,
        \___/                        the link after merge point M3 only has 3 units
    1   /M2                          of bandwidth available, so that the summed up
S4-----/                             reservation of 4 cannot be accepted.  
                               
The question is what do you do next, and the problem is somewhat similar to the
KR problem except that now you don't have a clear rule to determine who should be
denied its reservation in order to let the rest proceed.

You mentioned QoS routing a la PNNI to help out with this, and while I agree that
one can take advantage of the link load information that a protocol such as PNNI
makes available, you still need a mechanism to coordinate the requests from S1 to
S4.  And that is beyond what a protocol such as PNNI does today.

Hope this clarifies the problem I was pointing to.

Roch