[j-nsp] IS-IS multi-topology flaps when adding IPv6

Lars Erik Gullerud lerik at nolink.net
Thu May 19 13:33:51 EDT 2005


Hi list,

We have observed that when adding inet6 addresses to an existing IPv4-only 
link (Junipers on both ends), the following type of behaviour can be 
observed from IS-IS:

May 19 01:27:44  br1.stcy rpd[2845]: RPD_ISIS_ADJDOWN: IS-IS lost L2 
adjacency to cr1.osls on so-1/1/0.0, reason: MT Topology Changed
May 19 01:27:44  br1.stcy rpd[2845]: RPD_RSVP_NBRDOWN: RSVP neighbor 
X.X.X.X down on interface so-1/1/0.0, triggered by IGP neighbor down 
event

(the second line is not IS-IS related, but only included to show that it 
is indeed a full IGP adjacency flap, not just a cosmetic log insert)

This behaviour, namely that IS-IS drops the adjacency, seems to happen 
regardless of whether multi-topology IS-IS has been enabled or not 
(which, in our case, it is - with "topologies ipv6-unicast" set under 
protocols isis).

Now, I don't claim to be an IS-IS expert by any means, which is why I'm 
hoping someone can enlighten me on this - should it really be necessary 
to drop the L2 adjacency to establish a new protocol in IS-IS?

IIRC, IS-IS transmits "Protocols Supported" and "Interface Address" TLV's 
with every IS-IS Hello, so in my head, IPv6 NLPID TLVs and IPv6 Interface 
Address TLVs could simply be added to the next outgoing Hello-packet and 
the neighbouring router should be able to pick up the new protocol, and, 
if supported at both ends, start flooding new LSP PDUs for it.

In fact, rereading old papers (and RFC1195), I believe a router has to 
advertise "protocols supported" identically on ALL interfaces, and should 
therefore have the IPv6 NLPID TLV already present in it's IS-IS Hello 
packets if the router is IPv6-enabled - so wouldn't it then only need to 
exchange Interface Address (which are the link-locals on the interface 
I believe) and launch an SPF on only the appropriate topology once all 
necessary PDUs are exhanged, without affecting the others?

If someone can explain to me what part of the picture I'm missing (again, 
my IS-IS knowledge unfortunately leaves a lot to be desired), please chime 
in - since I gather Juniper would not have done it this way if it was not 
required, I would very much like to understand this better. :)

/leg


More information about the juniper-nsp mailing list