The Funet experience involved some bleeding edge IPv6 deployment, leading to a subsequently stable service.

In Q2/2002, the minimum IPv6 working version 5.2R2 was installed on the Juniper routers. Later, versions 5.3R2, 5.3R3, and 5.4R3 were also used. Version 5.6R3 fixed a problem with IPv6 loopback access list breaking Neighbor Discovery, but since then (as of Q1/2005) no IPv6 issues have come up.

The IPv4 network used OSPFv2 and BGP for routing, but OSPFv3 for IPv6 was not ready. Funet didn’t want to change the routing protocol, and for clarity, they wanted clearly separate IGP’s for IPv4 and IPv6: OSPF and IS-IS. In addition, if IS-IS had been deployed for IPv4 and IPv6, multiple topologies would have to have been supported as IPv6 routing would have been different (in parts) from IPv4. So, IPv6-only IS-IS was clearly the best (and only, discounting RIPng) choice at that time.

Unfortunately, there were problems in the Juniper and Cisco devices with IPv6-only IS-IS. For example, the Juniper platform would not support IS-IS only for IPv6. The first, and worst, problem was that the Juniper routers would always also advertise IPv4 addresses used in loopbacks and pointto- point links. Cisco’s IOS, unless you enabled IS-IS for IPv4 too (which was a non-starter for Funet), would discard all such attempts to form adjacencies: a total inoperability problem. Cisco’s IS-IS implementation has an option ‘no adjacency-check’ to override this; however, an undocumented fact was that it would only work (at least in this case) when using level-2-only IS-IS circuit-type (which was not the default). A first step in interoperability was gained when these were enabled in IOS.

Some problems continued. IS-IS route advertisements from Ciscos to Junipers were accepted in the route database, but not put to the Junipers’ routing table: this was caused by the above mentioned problem with adjacencies; this was reported, and fixed, in 5.2R2; a minimum workable version for Funet to use. The issue with Juniper always advertising IPv4 addresses in IS-IS was fixed, as (then undocumented) feature ‘no-ipv4-routing’ in 5.4R1. Also, one could not redistribute static discard routes to IS-IS (to generate a default route) until this was fixed in 5.3R3. You also could not set a metric when advertising a default route other than by redistributing a static discard route and applying a route-map in Cisco.

Fortunately, the rigorous tests in the lab network were enough to expose all of the above problems, which were fixed.

Also in 2002, most of Funet’s Cisco routers were replaced by Junipers in an upgrade of the network.

In early 2003, Funet noticed significant bandwidth bottlenecks (in the order of dozens of megabits/second) with their remaining Cisco equipment - 7200’s, 7500’s, GSR’s - regarding IPv6 forwarding capabilities, but the situation improved tremendously when software supporting CEFv6 come out later in 2003. This is very important for Funet as the IPv6 traffic level is at least 30-40 Mbit/s.