7200emu.hacki.at

[ccie14745]

Here's the version 3 of my MPLS lab. I've taken the bold step to include the LDP version (v3) along side the MPLS-TE version (v3.5). Both labs are identical in every way, except the method of distributing labels (and the TE tunnels). The bonus in this new version is the addition of IPv6 VPNs, using Cisco's 6VPE feature in 12.2.33 SRC. The VPNv6 setup is nearly identical to VPNv4, with the crazy addresses to throw us all for a loop. I was pretty amazed at how similar the configurations were. The VRF definitions were a change, but not a bad one.

There are a couple of caveats to note:

** The .NET file runs 12.2.33 SRC with the BARE MINIMUM memory (144MB by trial and error). I didn't get more grainular than 16MB, but 128 simply didn't work. You may notice messages about "reserving memory" on the router consoles. I suspect increasing the RAM per device would help this, but I didn't notice any crashing due to lack of memory.

** You will receive errors about reading stuff from flash or nvram or the like. I didn't bother to initialize anything as part of this, so the messages should only occur at bootup.

** Beware if you save your configs after changing them. I consistently get errors if I leave any of the "exit-address-family" commands in the configs, which I think is an issue with 12.2.33 SRC.

** There may be another bug in 12.2.33 SRC, but with the IPv6 BGP exchange between PE and CE. Many times, after initializing all the routers, I found that my IPv6 routes only propagated to one of the two CE routers in a VRF. Typing "clear ip bgp * soft" on the CE that didn't receive routes fixed the problem, but this only happens with IPv6. I didn't observe the behavior with IPv4 prefixes. I don't suspect the CE IOS version (12.3.21) to be the culprit, since that version has been really really stable for me in all my labs (and it has a great idlepc value and low RAM ceiling), but it wouldn't bother me if it turns out to be wrong.

** Using Dynagen 0.11.0, I could pretty predictably crash the hypervisors by loading the P and PE routers too quickly, one after another. I would watch your CPU % to become pretty steady before starting the next router. Not sure if this is a symptom on 0.11.0, but I haven't tried the 0.2.7 stable Dynamips yet to compare. Just be aware of loading the routers too short an interval between them.

** With everything loaded, I was at 60% CPU on my Q6600 with just under 1GB of RAM used system wide, so don't say I didn't warn ya. The lab could be loaded with just P1, PE1, and PE3 (and the CEs) to see the functionality work.

As always, the ZIP file has the necessary configs, .NET topology, and a PDF of the diagram I built. All in all, it was pretty fun to put together. Given the eminent shortage of IPv4 addresses for allocation, we will all become better with NAT, IPv6 or both in the next couple of years.

Good luck...
--
Justin White
CCIE #14745

[jumbo]