> The thing is, it seems like this (IMO relatively complex) iterative > comparison algorithm must be run for every outbound packet (one might be > able to optimize a bit with connection-oriented protocols).
I think connection oriented protocols effectively require that the source address (or addresses in the case of SCTP) be selected when the connection is created. Thus trying to do this for every packet would be the wrong thing in that case. For connection-less protocols where the application doesn't specify the source address in the bind() call, performance can benefit from caching the source address selected for a given destination much the same way that performance benefits from caching routing information for a given destination. > An alternative seems to be to implement some form of (unspecified -- > caching is only discussed in the context of dst address selection) > optimizations. One example of optimization could be putting the > to-be-used source address in the routing table, and refresh it always when > rtable or any address of the node changes or changes state > (deprecated/preferred, home address etc.), but these might have their own > problems. > > My worry is, is it useful to specify a mechanism for selection default > addresses that can't really be used without critical optimizations? I'm not worried about this. We have exactly the same issue for routing table lookup in hosts and routers; there is no specification that states how to implement caching schemes. Erik -------------------------------------------------------------------- IETF IPng Working Group Mailing List IPng Home Page: http://playground.sun.com/ipng FTP archive: ftp://playground.sun.com/pub/ipng Direct all administrative requests to [EMAIL PROTECTED] --------------------------------------------------------------------
