wet-behind-the-ears whippersnapper seeking advice on building a nationwide network

Owen DeLong owen at delong.com
Tue Sep 20 18:54:54 UTC 2011


On Sep 20, 2011, at 5:01 AM, Jon Lewis wrote:

> On Mon, 19 Sep 2011, Matthew Kaufman wrote:
> 
>> On 9/19/2011 6:02 PM, Jon Lewis wrote:
>>> On Sun, 18 Sep 2011, Frank Bulk wrote:
>>>> I should have made myself more clear -- the policy amendment would make
>>>> clear that multihoming requires only one facilities-based connection and
>>>> that the other connections could be fulfilled via tunnels.  This may be
>>>> heresy for some.
>>> That's not multihoming.
>> 
>> Really? Lets try these and see how you do:
> 
> The ARIN NRPM actually defines it:
> 
> 2.7. Multihomed
> 
> An organization is multihomed if it receives full-time connectivity from
> more than one ISP and has one or more routing prefixes announced by at
> least two of its upstream ISPs.
> 
> IMO, "full-time connectivity" would mean a leased line, ethernet, or even wireless connection, but not a GRE or other tunnel (which is entirely dependent on other connectivity).
> 

Why would you say that a GRE or other tunnel is not full-time connectivity? I have full-time GRE tunnels to two ISPs and they do actually constitute multihoming under the ARIN interpretation of NRPM 2.7.

> i.e. if you have a leased line connection to ISP-A, and a tunnel over that connection to ISP-B, and either A or your leased line fail, then you're down.  That's not multihoming.
> 

In my case, I have full-time circuits to two entities that provide very limited IPv4 services. I use those two connections to route GRE tunnels to routers in colocation facilities. My AS consists of the routers in the colocation facilities combined with the routers at my primary location and the networks to which they are attached. The GRE tunnels provide OSPF and iBGP routing to the routers at my primary location and my prefixes are anchored on the routers at the primary location. The colo routers provide the eBGP border connectivity to the upstream routers at each of the colos.

In what way is this not multihoming?

Now, let's look at some alternatives...

If I have only a single router at my primary location, is it still multihoming? I would say yes. Perhaps less reliable, but, that is not ARIN's concern.

If I have only a single physical link over which the multiple tunnels are connected, am I still receiving full time connectivity from two providers over the multiple tunnels?
Yes, actually, I am. Again, it's not as reliable, but, reliability is not ARIN's concern.

> Some of the scenarios you suggested are pretty unusual and would have to be considered on a case by case basis.  i.e. a shared T1 to some common point over which you peer with 2 providers?  I'd argue in that case, whoever provides or terminates the T1 in that case is your one transit provider, and again, you're really not multihomed...unless its your T1 and your router at the remote side, and that router has ethernet to the two providers...then that router is multihomed, and though most of your network is not, I'd argue that you have satisfied the requirement for being multihomed.
> 

I think you are delving much deeper into the internals of someones network than it is customary for ARIN to do in order to pass judgment on whether or not it is multihomed.

Owen





More information about the NANOG mailing list