> Routing replies  
> actually doesn't require involvement of the DHT (although some DHT  
> algorithms do 
> learn/cache based on reply paths). 

I don't think so. it is the case if there is no churn in the network. If there 
is churn, the reverse connection may not be available while routing replies. 
One of the reason is the neighbor peer in the connection table leaves the 
overlay or the connection is broken at that moment. In this case, how does the 
RELOAD process? only give up the transaction or fall back to the overlay 
routing based on peer ID?

On the other hand, essentially, I think, the forwarding layer introuce flexible 
routing decision, not only based on the overlay routing, but also reuse the 
established connection. IMHO, in order to improve performance, some new 
flexible routing decision may be developed  later in a generic way like what 
connection table does. We should keep the extensibility. 


_______________________________________________
P2PSIP mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/p2psip

Reply via email to