Hi Adrian,

> First: I think this is good work.


Thanks, much appreciated.


> I don't know how we document this to
> experiment with it. I suspect it needs some pretty sophisticated simulation
> because the rate of deployment is going to be such that the operators will
> not want to experiment in live networks. It seems (to me) to be in scope for
> RTGWG and is something that we should seek to stablise.

We would really need some realistic basis from one of the satellite operators, 
otherwise, we’re just making up garbage.

> Have you thought about how well the TE mechanisms will scale? Although the
> orbits are well-known (plus or minus meteor strikes), and although (as you
> say) a ground station only needs to know about its overhead stripes, and
> although you are not considering transit in this architecture, it seems to
> me that the computations will be more complicated being time-based, and I'm
> not sure that they are easily stable when pre-computed. 

One approach might be to compute the TE on an abstraction of the topology. For 
example, suppose that the topology can be roughly characterized as a grid. 
Flows can then be optimized onto an abstract grid and then satellites can be 
assigned process through grid positions. This saves total ground-zero 
recomputation.

> I also wonder, given how constrained ISLs may be, about the need for
> "centralised" computation to ensure that load is properly balanced.

I don’t see much alternative. The gateway has the topology and the traffic 
demand.  While the gateway doesn’t need to be a single CPU, distributing the 
data much more widely doesn’t seem advantageous. Doing path computation in the 
satellites itself seems like it would require some very expensive resources.

T

_______________________________________________
rtgwg mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/rtgwg

Reply via email to