Stupid may still be better

By: Scott Bradner

In an article in the September 13, 1993 issue of Forbes ASAP George Gilder wrote: "In the old world of dumb terminals - whether phones, IBM displays or boob tubes - a network had to be smart. .... But in the emerging world of supercomputers in your pocket or living room, networks will have to be dumb bandwidth pipes. (http://www.seas.upenn.edu/~gaj1/metgg.html) The idea that dumb networks are sometimes better than smart ones was examined and expanded by David S. Isenberg in his February 1998 ACM networks article "Rise of the Stupid Network" (http://www.rageboy.com/stupidnet.html).

The extreme example of a "smart network" is today's telephone network -- the self-labeled "Intelligent Network"(IN). The Internet is a quite good example of a dumb network.

The telephone network has evolved over many years to become a very complex collection of functions interconnected with an out-of-band control protocol called Signaling System 7 (ss7). Ss7 was originally defined to permit telephone network components from different vendors to talk to each other but local variants have developed in different parts of the world defeating the original purpose. The most basic assumption underlying the telephone network is that the device at the user's location is very simple. A telephone is merely a device which can do simple signaling, to request a connection to a specific phone number for example, and can translate sound into electrical waveforms or translate waveforms back into sound. Some phones have some additional functionality such as being able to remember common phone numbers, but these extra functions are not required to make a phone call. In other words, a telephone is a dumb device and the telephone network has to do all the work and, thus, has to have the intelligence.

Telephone switches have evolved into very complex and expensive systems. A typical telephone network does not have many switches. Many thousands of telephones can be connected to a single switch. In addition, a telephone network must have many other systems to do specific functions such as free-phone number translation, accounting, and billing. The networks that interconnect these systems are designed to be very good at supporting voice traffic. Specific data paths with defined latency and bandwidth characteristics are set up through the network to support each phone call. These paths are torn down at the completion of the call but, for the duration of the call, each system along the path reserves network and system capacity for that individual call.

At the other extreme, datagram networks such as the Internet are not nearly so complex. Individual datagram switches, known as routers, are quite small and, compared to telephone switches, very inexpensive. Datagram networks are hierarchical so that user devices connect to a local network, that network is connected to a network with a wider coverage, which in turn is connected to still another network with a still wider coverage. Instead of connecting tens of thousands of user devices to a single switch, each router in a datagram network will connect only a few networks, each of which usually have less than a 100 user devices connected to them. A typical datagram network will have many times as many routers as a telephone network would for the same number of user devices but each router is far simpler and less expensive than a telephone switch. But the most important feature of datagram networks is that they do not require large specialized systems imbedded in the network in order to function. The largest system in the Internet is the Domain Name System and that is actually made up of thousands of small computers distributed throughout the network and run by the organizations whose data they contain.

In a datagram environment applications, such as a new way for users to interact, are created and installed by the users -- the network is designed to support an endless variety of applications by not being concerned with just what applications are being run. Compare this to the telephone network where new applications must be developed by the telephone company and installed in the telephone switches. New applications can be deployed between developers on the Internet in minutes just because it seemed like a reasonable thing to try where it can take years for a telephone company to become convinced that a new application is one that is worth installing. Then it can take additional years for the application to be developed and installed in enough of the telephone switches for any two users to be able make use of it.

Those people who are in the telephone business point out that there are limitations to datagram networks - the main one mentioned is the difficulty of getting any level of service guarantee such as a maximum end-to-end latency, which is quite important for real-time interactive applications such as telephony. This is quite true in times of congestion but one factor that needs to be considered in evaluating the reasonableness is the fact that the bandwidth available for the Internet is growing dramatically in most parts of the world. There are still significant problems on the intercontinental links but in many countries new technologies such as wave division multiplexing, ADSL and cable modems are expanding the available bandwidth. In addition, the IETF is currently working on a new differential services set of standards which will allow a user to request different types of service, such as low-latency, from the network.

An example of where the dumb network vs. smart network conflict is clearly shown is in the area of Ethernet-enabled telephones. The traditional telephone people expect that Ethernet-enabled telephones will connect via a network to a large telephone switch and the switch will in turn connect to the destination telephone. Those people working with Internet technology assume that one Ethernet-enabled telephone will place a direct call over the network to a 2nd Ethernet-enabled telephone without going through a telephone switch. I predict that it will become clear to everyone but the most die-hard telephone person that smart networks cost too much to install and operate. The era of the dumb network is dawning.