[Outages-discussion] Internet "backbone"

[Patrick W. Gilmore]

On Nov 19, 2011, at 5:32 PM, Jimmy Hess wrote:
> On Fri, Nov 18, 2011 at 2:57 PM, Joseph Jackson
> <jjackson at aninetworks.net> wrote:
>> backbone?  This isn't the NSFnet anymore with a few long haul links between
>> sites.  When someone says Internet Backbone what are they trying to describe
>> since such a physical infrastructure doesn't really exist anymore.
> 
> It's true it's not NSFnet anymore,  but there still are some _much_
> higher capacity networks
> that are utilized for the greatest amount of traffic   from point A to
> point B,  and enough of those
> had an outage, the multitude of POPs and other interconnections simply
> would not come close to replacing them.

Of course some networks are bigger than others.  Of course some of the smaller networks buy transit from some of the bigger networks.

But it is not strictly hierarchical.  For instance, there are some networks who buy transit which have more traffic than some transit free networks.  And if two transit free networks lose connectivity, the traffic simply stops, so there is no worry about other interconnections or POOPs "replacing them".

Trying to simplify the Internet into a paragraph is not a good idea.  Trying to generalize connectivity on the 'Net at all is unlikely to be useful when speaking to engineers & architects.  We are not politicians or end users, you should not attempt to explain things in a way we know is more wrong than right for the sake of simplicity.


> What do you suppose the internet starts to look like when you consider
> only connections between networks and POPs where the connections have
> 100 gigabits/s or greater worth of aggregate connectivity?

It would not look like the Internet.  How would the graph of all airline flights look if you only looks at flights between hubs?  That would actually more closely resemble air traffic than your suggestion would resemble the Internet.

Oh, and 10x10G just ain't that big any more.  Which tells you something when you consider the fact there are orders (plural) of magnitude more "POPs" without it than with it.  For instance, even end user broadband networks have > 10x10G on their backbone in the US, while in many countries the international transit providers are lucky to have 1x10G on their backbone.  The large exchanges in Europe have individual LAGs > 10x10G.  While most IXPs around the world have < 10G total traffic.  Etc., etc.

Your suggestion makes "the Internet" look like a few lines between the largest cities in a few dozen countries and nothing else.  Not particularly useful, IMHO.


> Despite all the capacity and connection points, the network still has
> some logically  tree-like characteristics.

And far more characteristics that do not look like that.


> Inter-domain routing is specifically tree like.... with a small number
> of central  "Tier 1" networks   that form a logical backbone of the AS
> graph.

You are confused.  The Internet has not looked like a tree with the "roots" being the default-free zone for years.

I've said it before, I'll say it again: Most traffic never touches the "tier 1" networks.  This is not assumption or guesswork.  Even Arbor's inter-AS graph shows this, despite missing a great deal of traffic which does not touch the "tier 1" networks in their measurements (i.e. their graph has a bias towards the %-age of traffic touching the tier ones).

There is no backbone.  Hasn't been for a long time.  Please stop trying to find one in the spaghetti that is the Internet.

-- 
TTFN,
patrick