Older: 
IP: Unifying Force or Mask of Complexity?
Location: Flying to Boston
As we began our focus at Nortel on addressing the reality of the hyperconnected world, it became obvious that in addition to all of the fantastic gains that will come from being more connected, wherever you are, and via whatever device or application you need, that there will also be a cost. That cost will mostly be seen in the added complexity of living and operating in a world where an enterprise might have a million nodes to manage and billions of sessions to support. It might also be a world where the Internet is measured in trillions of devices and sessions. Because we, as the IT industry, have never before operated at that scale, we need to consider every aspect of IT system complexity and begin addressing it now, in advance of the scale and complexity challenges that inevitably are in front of us.
One area that we looked into at Nortel was the complexity of transport technology. When you look at the network today, what you see if you are an applications person is a “cloud” that provides best-effort Internet Protocol access to your own networks and the Internet. That’s a pretty good accomplishment because without IP we would lack a ubiquitous layer of addressing and protocol definition, and global communications would be difficult, if not impossible.
We could argue that IP has been a huge simplification technology and we would probably be correct. The issue, however, is that IP is not, in and of itself, the transport network. Underneath that nice, uniform IP layer is a host of diverse and complex networking systems that lack commonality; vary widely in cost, complexity and speed; and introduce huge complexity into the system.
If you consider why most enterprises don’t even consider running their internal applications over the cellular networks of today, it is not because those networks are unable to transport IP. It is because those networks transport in very different ways than their LANs, resulting in less speed (144 kb/s versus 100 Mb/s), greater cost (cost per bit in the LAN is almost 0; cost in cellular is not), different intelligence (security, provisioning and QoS), and lack of control. As well, so not to pick just on cellular, if we look at the differences underneath IP in the LAN versus the WAN, in Wi-Fi versus wireline, and in the metro versus the LAN versus the WAN… what we find underneath the nice uniformity of IP is that the systems we rely on today to actually deliver the traffic are fragmented, diverse, in many ways incompatible, and very costly.
What should we do about this? My suggestion is that we begin to look at this lower layer and ask if it is possible to converge to some common lower-level transport technology across all next-generation networking technologies. If so, what should that look like? Is there already an emerging dominant technology model that should be the goal of everything from the broadband wireless world to LAN/WiLANs, to MAN and WAN technology?
Below is a chart that I have used to suggest that this convergence is not only possible but is already happening.
The top portion shows the diversity under the IP layer today in each area (incomplete I am sure). You see circuit-switched and packet-switched technologies, slow-speed and high-speed links, high-cost and low-cost systems, and a huge diversity that makes end-to-end transport for all that IP traffic pretty inconsistent.
The bottom portion of the chart suggests that in each of these domains a convergence to “Ethernet-like” transport is already happening and that convergence is creating a “clear pipe” for the IP traffic to flow. In the cellular world, 4G wireless technology, such as Mobile WiMAX, is based on a model that looks a lot like Wi-Fi and Ethernet LAN systems in that they are packet-based, have high capacity, are low cost and support an open edge so that a wide range of devices and applications can operate over these networks without added complexity.
In the MAN and WAN, there is a set of technologies around Metro Ethernet emerging, where the optical network provides an Ethernet MAN service using Ethernet directly over optical systems. Nortel has been advocating this model with our Provider Backbone Bridging technology, and carrier uptake, while early, has validated 60-80% savings over legacy models of operation. In the WAN, the shift to Ethernet has been slower but the shift to packet transport has already begun (away from circuit) and, inevitably, there will be expansion of Ethernet in the WAN given the emergence - and now validation - of this model in all other networks.
This chart is a subjective view and the timing could be over 10 years but, logically, if we simplify the underlay of IP systems we will drive consistent services and lower opex and capex. All this is needed to reign rein (thanks K. Ramesh) in the cost and complexity that hyperconnectivity will create.
One last point to make on this topic… I want to preempt the dialog where some may suggest that we simply deliver IP over a physical media and skip a common packet transport layer below it. In situations where the applications and devices are intelligent and able to be provisioned, this model (IP over optics, for example) may be correct. The hyperconnected world, however, is far more complex than that.
Many of the new consumers of network services (sensor networks, for example) cannot support the overhead of a full IP model and thrive in simplified self-configuring network systems like Ethernet. Additionally, many of the tasks that make networks secure and stable happen in advance of entering the IP layers. For example, the current accepted model to control access via the authentication of end systems is based on Extensible Authentication Protocol over Ethernet (EAPoE, or 802.1X) because the ability to exchange credentials and determine trust in advance of interacting with the Internet has been shown to be a simple and very robust way to control access for a wide range of end points and systems.
This debate over whether the answer is IP over a physical layer, or IP over a common packet transport, is a good one to have, but the subject for another blog entry. The subject of this post is more to dialog over the need to simplify the layers under IP as a way to create the “clear pipe” and to drive down the cost and complexity of the hyperconnected network.
