RE: Minutes / Notes

["Tony Li" <Tony.Li@procket.com>]

|    
|    > So, i guess that you need to carry the locator of the 
|    destination, since it
|    > is needed to forward the packet to the destination.
|    > I guess that you donīt need to carry the destination 
|    identifier in all
|    > packets.
|    > For the source endpoint information, i am not sure.
|    > I think that carrying the source identifier would make 
|    more sense, since it
|    > identifies the other endd of the communication. 
|    
|    For the source part (locator vs. id) we need to understand 
|    how multicast
|    routing would work. Today it applies RPF to the source 
|    address and uses the
|    hierarchical structure of the source address to aggregate 
|    the information used
|    by the RPF. If the source identifiers are not aggregatable 
|    this will be an
|    issue.


For multicast, the source address needs to be the locator, NOT
the identifier.  RPF is a topological computation.


|    The ability to return packets without much overhead, such 
|    as an ICMP error or
|    a  TCP SYN, might be important to avoid a class of DoS 
|    attacks om routers.


Important, yes, but not because of DoS effects.  Just simple
rate-of-return arguments suggest that routers will do a better
"best effort" job of returning errors if they don't have to jump
through hoops.  And unlike hosts, the router cannot maintain an
effective cache of all of the sources that might send it erroneous
packets.


|    Having the source locator in the packet
|    means that an ICMP error can be generated without performing
|    any ID->locator mapping


Concur.  IMHO, this is a pragmatic necessity.

Tony