2011/5/8 Andrew Lunn andrew@lunn.ch:
Hi Daniele
It is not a real Link State routing, the document maybe is not completely clear. Every node before the OGM rebroadcast, attach its current best path toward the originator node. I could do this without adding an hop list, but simply substituting the TQ with PCE.
How you tried this? How well does this work?
I've not tried this, but it is perfectly equivalent as appending the current best path to OGM. I permit to the next node to do the same operation that previous node has done.
Not adding path information will be a big help in avoiding routing loops. It seems to me, just the simple idea of weighting the local TQ with a bitrate factor might help.
Routing loops can occur exactly as they can occur with current protocol, I think. I did not introduced any cause of loops, I use the list of hop received in OGM only for PCE computation. Anyhow I will think deeply on this potential problem.
The hard part is correctly handling the total air time. One 36Mbps hop is better than two 54Mbps hops, because of the total air time.
I explained this in the report. I did not calculate the exactly total air time, but I weight the metric with its length when comparing it with the current best route. For example, for doubling the length I require 4 times more PCE.
It would also be interesting to see how bit rates change with usage. It could be that you pick a next hop partially based on its bit rate, which puts load onto the next hop, and its bit rate drops, since it is now being used more and so suffers more interference. The dropping bit rate causes the next hop route decision to change.
It could be, you are actually introducing more route flipping.
Certainly bit-rate is subject to frequent oscillations, so I maintain an exponential moving average of PCE that "smooth" this oscillations. Bit-rate is actually subject to variation for various factor, such as node distance, channel occupation, etc.. and all these factor are important for routing decision. I take the bit-rate that summarize all these information and use it to weight the TQ.
What would be interesting is to look at the effect of route flipping. Is it bad for performance? If so, can hysteresis be added without causing routing loops. There was some work posted here a couple of months ago that started to look at these questions. Unfortunately, it did not consider if routing loops would be introduced.
I've seen that work and seems interesting. Routing loops have been not considered yet in any published paper as far as I know. :(
Andrew
Thank you for your comments!