Franklin F. Kuo

At the invitation of the China Institute of Communications, the Citizen Ambassador Program of People to People International arranged a visit for a delegation of professionals in telecommunications and networking technology to the People's Republic of China. I was the leader of the delegation whose purpose was to exchange information and solidify contacts with Chinese professionals within the computer and communications industry. The exchange focussed on topics dealing with China's telecommunications infrastructure, especially on subjects relating to current research and applications in computer networks. This group, known as the Networking Technology Delegation, consisted of 24 members, with 18 professionals and 6 accompanying spouses. Of the 24 there were 17 people from the United States, 5 from Japan (2 U.S. citizens residing in Tokyo) one from Germany, and one from Hong Kong, so it was truly an international group. Six members of the delegation were native Chinese speakers, which made it very convenient in our travels.

The itinerary for our trip was as follows:

May 11-15, 1992 Beijing May 15-19, Xi'an

May 19-22, Shanghai May 22-24, Hong Kong

Our professional visits concentrated on telecommunications and on computer networking. We visited a number of the major research institutes and universities in Beijing, Xi'an and Shanghai. In all three cities a lot of interest in computer networking was evident, especially in the Internet and how to connect to it. Another question that frequently came up was the future of OSI vs that of TCP/IP. There are a lot of local area networks in operation in China, which connect many PCclones, mostly Chinese produced, to some older generation mainframes such as Honeywells. What we did not see were operational wide-area networks (WAN), with the exception of a metropolitan demonstration network in Beijing.

Current Wide-Area Networks

At present, the major WAN in China is the China National Public Data Network, CNPAC, that is currently being developed and implemented. CNPAC, an X-25 packet switched network, is designed to carry data at speeds varying between 1.2 and 9.6 kbps. The hub is in

Beijing, where the network management center is located, with packet switches sited in the major cities of Shanghai and Guangzhou, and PS concentrators found in other major cities. The packet switches, concentrators, and PADS (packet assembly/disassembly devices) are all manufactured in China. At the Beijing hub, there is an international access line to CNPAC. Since we did not see a CNPAC demonstration, it is not clear how much of it is operational and how much is still under develop

ment.

Other private data networks are used in China in applications, in the railway system, banking system, and civil aeronautics.

In China, an X.25 link to the Internet uses a store and forward system via the CNPAC international access line in Beijing connecting to the University of Karlsruhe, Germany. To the outside world, this link is being called "the China Academic Network (CANET)." In addition to CNPAC connectivity, there is dial-up access to CANET from inside China. David Kahaner of ONR Tokyo reports that he frequently communicates with Chinese scientists via CANET. However, many of the Chinese networking specialists we talked to have never heard of the

name "CANET," so we suspect that CANET means more to the outside world than in China.

The major problem confronting the development of WANS in China is the poor telecommunications infrastructure. Since the penetration of basic plain old telephone service (POTS) is less than 1% among Chinese businesses and households, and since local and long distance telephone switching and transmission facilities are inadequate and antiquated, it is difficult to build a modern computer network upon the current telecommunications infrastructure. It will take decades to bring the basic telecommuncations system up to modern standards, so Chinese networking will take a long time to come up to western norms also.

Metropolitan and Campus Networks

In Beijing, we witnessed a very impressive metropolitan networking project called NCFC (National Computing and Networking Facility of China). NCFC is a demonstration network in Beijing that links the two major universities, Tsinghua and Beijing Universities to a number of research institutes of the Chinese Academy of Sciences (CAS). Each of the pariticipating institutions have campus networks like the TUnet of Tsinghua University. These campus networks are connected by NCFC as a two level system. Currently NCFC has a 10 Mbps backbone connecting the three campus networks that will increase to 100 Mbps in the next phase of the project. Communication protocols will be ISO/OSI, but TCP/IP is the first phase protocol. The top level of NCFC consists of the backbone and the network control center. The second level is composed of campus networks at the two universities and CAS. NCFC is the

largest and most ambitious networking project we saw in China. It is partially funded by the World Bank and the State Planning Commission, and is in limited operation now, with full operation expected by 1994. We visited two of the three groups that are participating in the development of NCFC. The first was the Computer Network Center (CNC) of the Chinese Academy of Sciences, with its own campus network, the CASnet. The CNC seems to have the major responsibility for the development of NCFC and is staffed by 40 professionals. The second group, at Tsinghua University, is described below.

The Tsinghua University Network (TUnet)

The most impressive university networking group we visited was at Tsinghua University, the premier technical university in China. Under the direction of Professor Hu Daoyuan, the Tsinghua University network, TUnet is being developed under a well laid-out strategy based upon the following goals:

1. It will be a universal,

comprehensive campus network; its usage will include instruction, research, administration, library, and communications services.

2. It will be a multimedia integrated services network; messages transmitted in the network will include not only data, but voice and video as well.

3. It will operate under accepted international standards for interfacing devices to the network. Emerging standards are important in the fast changing technology of networking. Ini

tially Tunet will operate under TCP/IP, but migration strategies have been adopted to migrate to ISO/OSI.

4. It will be a heterogeneous network that uses a variety of advanced networking technologies (LAN, PABXPS, ISDN, and FDDI) to interconnect multivendor computing facilities.

5. It will be developed in phases. The first phase (1987 to 1991) concentrating on interconnection of facilities, and the second phase (1992 to 1995) emphasizing network services.

In Tunet there are three major networking facilities:

a. A circuit switched network based upon an integrated services PABX,

b. A packet switching network based upon X.25 switches and PADs,

c. Ethernet LANS interconnected through a 100 Mbps FDDI optical fiber backbone.

A key function of TUnet is electronic mail. Tsinghua University's message handling system (MHS) funtions includes mail, telegraph, teletext, fax, videotex, voice, and images. The MHS is based upon the EAN system developed by the University of British Columbia conforming to the CCITT X.400 recommendation series of 1984. Tsinghua's work on its e-mail system includes migration, Chinese localization, menu adaptation and the implementation of remote user agents.

The work at Tsinghua on TUnet and NCFC underlines one of the basic constraints that Chinese net

working technologists must live with. Unless you have foreign (hard) currencies to purchase networking equipment, you've got to design and build everything from scratch, including hardware and software. So TUnet represents in many ways a bootstrap operation. The people in TUnet are all very well trained and dedicated. It is unfortunate that they could not make use of technology that is readily available in the Western world.

Local Area Networks (LANs)

In China today many LANs are in use. Two common LAN products widely available throughout China are Ethernets from 3 COM, and Netware, a LAN operating system developed by the Novell company.

These products are available in China because of joint venture arrangements that the cited companies have made with Chinese counterparts. Most of the LAN products are manufactured in China under license from U.S. companies such as 3 COM and Novell. At the Shanghai Jiao Tong University, Professor Yang Chuan-hou, the Director of the Computer Network Research Laboratory, presented to us some work ratory, presented to us some work that dealt with an architectural design of a gateway interconnecting LANs to an X.25 packet switched network. The work again was developmental in nature, in that both hardware and software designs were implemented in the laboratory.

Work like that of Professor Yang and his colleagues illustrates the practical nature of the network

ing development going on in China today. I visited China in 1984, when I was a World Bank consultant to Shanghai Jiao Tong University. In the intervening years, there has been an explosive growth in both computing and networking technology in China, which will only accelerate with the further penetration of the Internet into China. Since the Internet is capable of bringing network specialists and users throughout the world into a larger cooperating community, I believe that China's networking community will soon become full partner in this world-wide community.

by David K. Kahaner

INTRODUCTION

A transputer is an elegant little processor that is produced in a variety of versions by the U.K. company, Inmos (newest is T9000), with respectable computing power. (The T800 has about 2 MFLOPS, or 13 MIPS of peak performance.) On one chip there are a CPU, a floatingpoint processor, built-in support for interprocessor communication, and microcoded context switching capability. Transputers have been configured as boards for PCs and in large parallel configurations with peak performance of several hundred GFLOPS. Transputers can be physically connected by four interprocessor communication links. The communication is not exceptionally fast, but it can be performed simultaneously with calculation. There is also a unique language, Occam, that appears to be a combination of parallel C with process control.

The main interest in transputers appears to be that a user can put together a system of one or more and get it up and running very easily. The four physical links make a twodimensional mesh natural, but by software, other kinds of interconnections can be simulated, such as rings, pyramids, hypercubes, etc. Thus it is possible to experiment with a variety of parallel processing considerations, and several different companies pro

duce both general and special purpose parallel computers that are transputer based. Not unexpectedly transputer use is highest in the U.K. and in Europe, but user communities exist in many other countries. I have noticed that universities in Asia looking for inexpensive ways to move into parallel processing often develop systems from transputers.

Discussions with Japanese scientists suggest that transputer activities are widely dispersed in Japan, especially focused on applications. Nevertheless, within the computer science community, transputers are not at the center of attention, although this is false for those institutes that have emphasized work on them. (I would like to hear some opinions about that point.) Weaknesses center on performance degradation with scale-up to larger systems, performance a bit behind other systems at the same point in time (one scientist commented that the T9000 would have been impressive if it was available a year and a half ago), weak compilers other than Occam, and general complaints about the difficulties using Occam.

The Fourth International Transputer/Occam Conference was held in Tokyo during the first week of June 1992. There were two days of tutorials, followed by about a dozen and a half conference papers. One third of these were authored in the United

Kingdom or Europe and most of the remainder were from Japan. Approximately 100 scientists attended, perhaps 90 Japanese. Titles and authors of the papers are given in the Appendix. A proceedings was published and is available as follows:

Transputer/Occam Japan 4
S. Noguchi and H. Umeo, Eds.
Proceedings of the Fourth
Transputer/Occam International
Conference

4-5 June 1992, Tokyo Japan
IOS Press

Van Diemenstraat 94 1013 CN
Amsterdam, Netherlands
For other details, contact one of
the conference organizers.

Prof Shoichi Noguchi
Director, Research Center for
Applied Information Sciences
Tohoku University
2-1-1 Katahira Aobaku
Sendai, 980 Japan

Prof Hiroshi Umeo

Dept of Applied Electronics
Faculty of Engineering

Osaka Electro-Communication
University

18-8, Hatsu-cho, Neyagawa Osaka 572 Japan

Transputer meetings always seem (to me) a bit different from meetings about other parallel computers,