Pierre Bouriez <Pierre.Bouriez@multicable.fr>
Olivier Muron <Olivier.Muron@inria.fr>
Institut National de Recherche en Informatique et en Automatique
The paper describes MulticablE, an experiment in home Internet access through the Paris Cable TV Network. The terminals used are personal computers.
Keywords: Internet at home, high-speed access, cable modems.
Lyonnaise Communications is the leading French cable television network, with 550,000 subscribers. It is in the process of developing new services to exploit the full potential of its network. Based on such factors as the growth of households with PC equipment, the increasing demand for Internet access, and the emergence of new technologies that use CATV networks as data networks, Lyonnaise Communications decided in early 1995 to launch a field-test of high-speed online services and rapid Internet access using the CATV network.
This experiment had two goals:
The MulticablE experiment began in January 1995 and received an Information Highways and Services Program rating from the French government on 16 October 1995.
The program had three phases. During the first phase of the project (1995), technical and marketing aspects were studied. The second phase, the experiment itself, was conducted in Paris's seventh arrondissement, using a test set of 200 cable customers. It began at the end of 1995 and ran for six months. The third phase was data analysis. Data were collected during the second half of 1996. Participants were interviewed by telephone and traffic statistics data were analyzed. The results of the experiment showed the feasibility of opening a commercial service, and one in Le Mans is now planned.
In 1995, only a handful of vendors were considering delivering products for the emerging cable modem market. Selection criteria were actual availability of the products, which eliminated all vendors but two; immunity to noise transmitted on the return path of the CATV network; throughput performance; and cost of the system. Considering the fact that noise in the return path still was not well known at the time, immunity to this noise was the decisive criterion. Intel, which proposed a dynamic allocation of return-path frequencies, was the most efficient system, and was selected for implementation.
The technical study was conducted with France Telecom, owner of the physical network. It concluded that the Paris Cable TV network, a 860-MHz fiber-trunk/coaxial distribution network, with a return path of 25 MHz, could be easily adapted for interactive data services.
Ten fiber nodes exist in the test area, and each consumer is connected to them through the regular CATV coax. In this coax, 6 MHz out of the 860 available have been dedicated to data transmission from the nodes to the users, and about 2 MHz out of the 25 available have been dedicated for data transmission from users to the node.
The 10 nodes and the headed were organized as a FDDI loop, data at the nodes being transmitted over coaxes through modulators and extracted through demodulators. At the headend, an Ethernet LAN installed included a POP (a device managing frequency allocation, modem by modem, node by node), a proxy server, a mail server, and a Web server for proprietary content. From the headend, leased lines provided access to the Internet.
The architecture of the network and the services was defined with help from INRIA. TCP-IP protocols were chosen for the network, because of their openness, their performance, and the low cost of their implementation. For services, HTML was chosen because of its wide selection.
In early 1995, there were endless debates about the future of online services versus the Internet. It was decided to test them both, and to generate content that would emphasize the high throughput capacity of cable modems, while leasing high-speed lines to access the Internet.
The interface design was important. The interface would be the first contact most users would have with the Internet and online services, so it had to be user-friendly and devoid of technical and configuration terminology. Consultants came up with a village design in which each building represented a family of services.
A marketing study showed that PC penetration was very high among CATV subscribers in Paris: more than 40 percent of CATV households owned a PC at the end of 1994. The seventh arrondissement was chosen because of its high PC penetration and its large average household size. The demand for interactive services and Internet access was also growing rapidly. A group of services was chosen, taking into account network characteristics (high bandwith of the downstream channel). This included real-time audio and video transmission. The billing method used was a flat monthly rate that included unlimited access to the Internet.
Participants in the experiment were recruited through direct mailings. The response rate was 13 percent, well above the national norm. The two incentives offered were proprietary online service for Fr 50 per month and Internet plus online service for Fr 200 per month. Ninety percent of the respondents chose the most expensive offer, an indication of the value and expectations they placed on the Internet, and the perceived limitations of online services. A second mailing that targeted working women was sent in order to prevent under-representation of this population segment in the experiment. Price of the service was not a barrier for experiment recruitment.
The selection of the final 200 participants out of a pool of 700 candidates was made so that the majority were people who had never seen or used the Internet before, 30 percent were not CATV subscribers, and 5 percent were first-time home PC users. Ninety-five percent of the participants lived in households with PCs.
All participants had the experimental connection installed in their homes. Use of the service was unlimited, and users were able to surf all night long if they wished because there was no connection to telephone lines that would affect family use of the telephone. In part, this explains the interesting differences that were found in Internet usage, which we will describe later.
Initially, installation of the software and hardware needed for the experiment required two hours. Over time, the technicians were able to reduce this installation time to about one hour. A hot-line service was installed for user problems; it turned out that most calls were related to application software configuration (mail or browser), not to the network itself.
A study was performed using the proxy server's log file. An analysis was performed of the probability distribution of the times between successive requests of the same user. The corresponding histogram is shown below.
Analysis of these distributions led us to consider an inactive period longer than 1,000 seconds as a new connection. Using this definition, the average connection time was 55 minutes.
MulticablE use was daily and heavy. Use of the Internet services was very high (85%) compared to proprietary online service use. It was recreation-oriented. After eliminating search engines, browser company home pages, and other toolbox sites from the statistics, the top three sites were as follows:
These are the leading subject areas in pay-TV subscriptions in France. The hours of heaviest usage also coincided with TV-audience viewing patterns.
The "user-friendly" interface quickly showed its limitations; within a few days, users mastered search engine operation on the Internet and got bored with the interface design. However, they were grateful to have it, for it succeeded in dispelling the notion that the Internet was too complex, initially fostered by the French press.
The most common user was 35 to 50 years old. We noticed that, on an average, the number of users per PC increased by one during the test period. The service was used most frequently by two or more family members. The participants mentioned, among the advantages of this kind of access, speed, permanent connection, flat monthly rate, and availability of the telephone line for regular use.
Network performance was quite impressive. The maximum throughput of the network was limited mainly by the PC itself or by the source site. The noise impact, feared to be the most critical issue, did not appear to be troublesome. The only cases of annoying noise were generated by cable modem emissions far above specifications. This early problem was solved through software upgrades.
Technical problems occurred where they were not expected--in the home PC. Configuration of TCP/IP software and Ethernet cards could not be replicated from one user to the other and appeared to be unstable; users often lost their TCP/IP configuration while adding new features in their PC. Mass-market consumers were not interested in solving these problems themselves.
In the matter of the architecture efficiency, the FFDI loop interconnecting all-fiber nodes proved to be an expensive choice. Having many active elements in the nodes makes them more costly to maintain, and their centralization in the headend does not alter performance and simplifies maintenance procedures. It was decided that further development should include centralized architecture.
The test gave us a better understanding of the "frequency allocation vs. TDMA" debate now going on in cable modem technology. Allowing only one user per frequency led to a capacity problem from time to time, which resulted in having some users rejected from the system. This would not have not happened with TDMA. However, TDMA would have rejected everybody at the same time from the system only whenever noise occurred. The best solution lay in a compromise: allocating several frequencies in a range, and allowing several customers at the same time in each frequency allocated. This proved to be more efficient in the test which subsequently took place in Le Mans.
Preliminary results of the MulticablE experiment were considered very promising by Lyonnaise Communications. In September 1996, the Cable TV operator decided to open a commercial Internet access service in Le Mans, 200 km west of Paris. After conducting a technical analysis of the products available on the market, it chose CyberSurfer, made by Motorola. The bandwidth permitted by the newer technology is 10 Mbits/s downward and 768 Kbits/s upward. This commercial service is now open for private and business use. A local area network can be installed for business customers. Servers can be implemented for efficient consumer use. For home users, the cost is Fr 195 per month with unlimited length of connection. The opening of a Paris branch next year is under consideration.