Up until this point, I have considered some of the folklore surrounding the packet-switched Internet. The overall goal is to provoke discussion and research on fundamental issues that need to be addressed so that IP can continue to revolutionize the world of communications. As a research community, we need to think beyond the daily challenges of maintaining and optimizing the expanding Internet, and move on to consider the enormous challenges that lie ahead.
It seems that there are two main limitations to the widespread adoption of IP: dependability and the right way for IP to co-exist with circuits. In what follows, I will discuss each in turn.
High dependability, in the broadest sense, is a must if IP is to become a successful transport technology (to compete or displace circuit-based transport networks), and if the Internet is to become the universal infrastructure for high value applications. For example, voice services are a high-revenue, and very profitable business. Trusting them to today's unreliable, and unpredictable IP networks would be an unnecessary risk, which is why -- despite predictions to the contrary -- telephone carriers have not done so.
High dependability means several things: robustness and stability,
traffic isolation, traffic engineering, fault isolation, manageability, and last but not least, the ability to provide predictable performance in terms of bounded delay and guaranteed bandwidth (QoS). In its current form, the Internet excels in none of these areas. Although it is clearly a challenge to achieve each of these goals, they must all be solved for IP to become dependable enough for use as a transport mechanism.
The current Internet is based on packet-switched routers in the edges, interconnected by a circuit-switched transport network. Given the benefits of circuit switching, it would seem perverse for the packet-switched network to grow to subsume the transport network. It is inconceivable that the network providers would remove the existing, robust, reliable, predictable and largely paid-for transport network, and replace it with a technology that seems more complex, less reliable, more expensive and not yet installed.
What seems more likely is that packet switching will continue to exist at the edge of the network, aggregating and multiplexing traffic from heterogeneous sources for applications that have no delay or quality requirements. In other words, packet-switched IP will continue to provide a simple service abstraction for a variety of applications. However, this does not preclude the existence of highly specialized service networks living alongside IP and using other switching techniques. In fact, it is unlikely that the phone or TV cable service networks will be completely replaced by an IP network any time soon as it would require a huge amount of capital to build a new network.
At the core of the network, one can expect the circuit-switched transport network to remain as a means to interconnect the packet-switched routers and as a means to provide high reliability and performance guarantees. Over time, more and more optical technology will be introduced into the transport network, leading to capacities that (necessarily) electronic routers cannot achieve.
One remaining question is whether or not the circuit-switched network will be controlled by IP. In other words, will the IP network decide dynamically when to create new circuits between routers? For example, a router could monitor the occupancy of its queues or the number of active flows and periodically add or remove circuits to other routers based on current demand [7,181]. Such a system has the benefit of enabling IP to gain the benefits of fast optical circuit switches in the core, yet maintain the simple service model for heterogeneous sources at the edge.2.10
However, while a complete control by IP of the circuit-switched backbone seems appealing to IP, one needs to remember that the majority of the revenue for the circuit switches will still be from other applications, such as voice. Since the packet-switched network is unlikely to provide the predictability needed for voice traffic, it will continue to operate over its own, separate circuit-switched edge network and to be carried over the shared transport network at the core. In this environment, it is unlikely that the routers will be allowed to control the entire capacity of the transport switches, unless the revenue for the Internet exceeds that of telephony. At the current growth rates, it will take over 15 years for data traffic to surpass telephony as the main source of revenue in telecommunications. In the future, it is more likely that the routers will be allocated a fraction of the circuit-switched transport infrastructure, which they can control and adapt to best serve their needs.
With the dynamic control of circuit-networks (possibly by an IP-based control plane), it is also conceivable that the IP routers at the edge can signal to the transport network to dynamically create new circuits or change the bandwidth of existing circuits.
In the preceding discussion, an outcome was depicted based on historical conditions, in the context of a pre-existing circuit-switched transport network. So if we started again, with the benefit of hindsight, would we build a network with circuit switching at the core, and packet switching at the edge?
I believe that we would, and that it would look something like this:
Additionally, we will find more hybrid switches that can do both circuit and packet switching, serving as gateways between the two worlds. Chapter 4 and Chapter 5 describe two ways of bridging packet switching and circuit switching.
Finally, the idea of using circuit switching to interconnect distant routers can also be extended to using a circuit-switched crossconnect to interconnect the packet-switched linecards of a router.