The topmost RED LEVEL is commonly called the Internet "Backbone". It is the main highest-traffic network that interconnects the routers of the second level.

Internet Routers — shown as round black "nodes" in this diagram — examine the destinations of individual Internet data packets to determine the best route to use in forwarding the data to its final destination.

The BLUE LEVEL is generally composed of large "Tier-1" service providers. They purchase "backbone" bandwidth from the major communications carriers and resell it to smaller service providers, universities, and corporations.

The GREEN LEVEL is generally composed of smaller (non-Tier-1) service providers, universities, and corporations. They typically purchase bandwidth from Tier-1 providers and interconnect the individual machines within their organizations into a "Local Area Network" (LAN).

The BLACK LEVEL is composed of the connections to individual machines. This is often called the "Last Mile" of the Internet and ranges from dial-up modems, to ISDN, to DSL, to Cable Modem, to local Ethernet segments.

What's wrong with this simple system?

We all know from experience how incredibly reliable the Internet has proven to be. The original design, which was funded and roughly specified by DARPA — the U.S. Defense Advanced Research Projects Agency — specified that this "Internet" should be able to withstand and remain functional after an atomic bomb blast.

But returning to our first diagram above, you'll notice that simply "taking out" that main center node, joining the various red top "backbone segments", would instantly chop our "Internet" into four separate disconnected pieces. That's not good. And even absent an atomic bomb, technicians have been known to trip over cords or type in the wrong command, resulting in a loss of network connectivity. The network shown above could only work if every piece worked perfectly. In other words, this simplistic network design is extremely fragile.

It has another problem too:
This simple "tree-structured" architecture routes all traffic between top-level backbone segments through the center node. In other words, a large percentage of the TOTAL bandwidth of the Internet would be flowing through this node, and it's unlikely that any current technology on Earth could adequately handle that much traffic.

If you'll take a moment to examine this diagram you'll notice that the central router has been eliminated in favor of direct "peering" links (RED) among the top-level backbone routers. In cases where traffic demands it, duplicate links are established to deliver reliability and additional point-to-point bandwidth.

At the major Tier-1 ISP (BLUE) level, the ISP's have created redundant links between many of their own routers, and have even established "peering relationships" with other ISP's in order to share bandwidth and increase their mutual reliability.

As you can see, with multiple redundant paths to router nodes at every level, a great many links and/or routers can be "down" or disconnected at any time without isolating any region of the entire network. While the smooth flow of data from affected points might suffer a longer than usual transit, especially if lower capacity links were overburdened by a rerouting of traffic, the system "fails soft", degrades gradually, and the data can find a way to its destination.