EIGRP Question

Netstudent

The originating router does not consider the Diffusing Computation to be complete until replies have been received from all the neighbours. There is an Active Timer that has a default value of 3 minutes. This timer is used to time how long it takes to perform the Diffusing Computation. On a large network where a chain of routers may end up performing the Diffusing Computation, it may be a while before the originating router completes. If all the replies are not received within this 3 minutes, then the route is said to be Stuck-in-Active (SIA). The neighbour involved is removed from the neighour table and the metric for that route set to infinity so that another neighbour can meet the Feasibility Condition and become a Feasible Successor. If an EIGRP network is particularly large, or there are a number of low bandwidth links such that it takes a while for replies to get back, then those neighbours that have yet to reply have their Reply Status flag set. If no reply is received from a particular neighbour before the Active timer times out, then the neighbour will be removed from the neighbour table. If a reply DOES come back after the Active timer has timed out then the neighbour gets reinstated

OKay so I read this paragraph and I read from other sources that say if a router sends a reliable packet to a neighbor, and that neighbor does not send back an ACK, then the packet will be resent as reliable unicast 16 more times. But I read other places that a query is always multicast and a query is reliable.

So can a query be sent unicast if an ACK is not sent back by the recieving neighbor?

Secondly, if a query can be sent as unicast to a neighbor, and that neighbor has no feasible successor, will it then send a query to it's neighbors as multicast? IN this case the unicast query would only go to the neighbor and the rest of the diffusion will then be multicasted if the neighbor has to feasible successor to offer, thus the neighbor becomes active as well??

I think I might be getting confused between ACK's from a reliable transmission and replies to a query wich are are both reliable and need an ACK. I know that replies to a query are unicast so theres not any confusion as to whether they can be resent unicast. Any insight is appreciated.

dtlokee

1. Queries will be multicast to the neightbors and the router will expect an ACK (the timeout for the ACK is determined by the SRTT value) from all neighbors on the link that it knows of. If there is no response from a neighbor the query will be resent as unicast.

2. If a neighbor has a route it will send the route to the neighbor, if not it will forward quries to all of it's know neighbors, and so on. In large networks this is what can result in SIA when there are no neighbors with a feasable successor to the lost route but it needs to go all the way to the endpoints (an EIGRP router with no neighbors other than the one the query was received from) before the queries will end.

This is why stub was introduced, basically to tell the router not to send a query to that neighbor in the case of a lost route. This can also result in some odd behavior of EIGRP when you have a couple neighbors defined as stub and a route is lost. It is possible there is a path to the destination through one of the neighbors defined as "stub" but because the router cannot query the stub router it will never learn of the new path.

Netstudent

Thank you.