Re: [ns] Multipath Question

[Lloyd Wood <l.wood@eim.surrey.ac.uk>]

On Wed, 10 Oct 2001, Renato Azevedo wrote:

> I would like to know how I can send packages in different percentages
> for two alternative paths.
> I already used mutipath routing and I conclude that for paths with the
> same cost the flow traffic divides itself for the two paths in equal
> percentage. 

not quite. 

In the ns multipath routing implementation, all incoming traffic is
divided among the next hops of the equal-lowest-values paths between
the current node and the destination, in equal amounts on a
round-robin basis for each output packet independent of flow identity,
which is a slightly different thing.

Consider a topology of equal-cost links

    +-+-+-+-+->+B
    | | | | | 
A+->+-+-+-+-+

where each node (+) is using multipath routing, and so division of
incoming traffic from A to B takes place at each node. An increasing
amount of the traffic will end up flowing along the top line as we
travel to the right, which is unfortunate if you have decided your
same-cost paths are, respectively, the top and bottom lines joining A
and B.

> If  I modify the cost of one link, the flow converge to the link of less
> weight. 
> I want to be able to send for example 70% of the flow for one link and
> the others 30% for another link. How can I obtain this division of
> traffic?

by modifying the multipath classifier's C code, and getting it to
recognise individual links. Tying it into identifying and remembering
links can be tricky, though. So...

    +-+->+B
    | |
A+->+-+->+C

You could insert the multipath topology above in place of your single
node, using low-delay high-speed links, to give you 25% output (50% of
50%) at C and 75% (50% + 50% of 50%) at B. (It would be helpful if
multipath can be set on a per-node basis to allow more
fexlibility with the rest of your topology; I'm not sure of the
current state of that, and it has had fixes in the past.)

[I'll be happy never to do multipath simulations ever again; some of
 mine ran for >24 hours. the tight and inflexible schedule of the
 resulting IEEE Comms Mag publishing process was a pain, too - I got
 to correct proofs repeatedly during travel, on courses and even
 between parts of my PhD viva, constantly missing  deadlines due
 to timezone differences, despite warning them months in advance and
 asking to do proofing asap to avoid the possibility of this. I
 wonder what happens if you have the temerity to schedule a vacation
 at such a time?]

hth,

L.

<L.Wood@surrey.ac.uk>PGP<http://www.ee.surrey.ac.uk/Personal/L.Wood/>