Lapas attēli
PDF
ePub

VARIOUS COMPUTING ACTIVITIES

IN TAIWAN

The International Symposium on Algorithms '91 and visits to
Academia Sinica and the National Center for High Performance
Computing in Taiwan are summarized.

by David K. Kahaner

SUMMARY AND BACKGROUND

I first visited Taiwan, Republic of China (ROC), in December of 1990 [see my report "Computing in Taiwan," Scientific Information Bulletin 16(2), 23-29 (1991)]. The current report provides updates and further perspective. At the time of my first visit I commented on the lack of applications research. I still sense something of a disconnect between what appears to be very high quality academic research and industrial needs. However, in the area of computing, the soon-to-beopened National Center for High Performance Computing will go a long way towards bringing research and development (R&D) closer together.

Please refer to the earlier report for general background on Taiwan. Suffice to say that Taiwan's 20+ millions are enjoying a remarkable period of economic growth. The country's foreign exchange reserves including gold exceed $90B (greater than Japan's) and are projected to go beyond $100B in 1992, driven in large part by a trade surplus of about $13B in 1991. Per capita income, $8,800 (third highest in Asia behind Japan and Hong Kong), is growing about 9% annually, as is general economic growth in spite of the

world-wide recession. The savings rate has been hovering slightly over 30%. Since 1971 the country's gross national product has gone from $6.6B to $180B. Europe is a huge trading partner, too: $14B in 1991, with a trade surplus of over $4B. In addition to trade with the West, there is a growing volume of trade with mainland China, up nearly 50% last year (1991). So-called crossStrait trade was more than $4B for the first 10 months of 1991. Recently there have also been calls for a loose economic community between Taiwan, Hong Kong, and the coastal provinces of southern mainland China, perhaps even including Singapore.

A government-backed consortium led by the Taiwan Aerospace Corporation (TAC) has proposed to purchase 40% of McDonnell Douglas' commercial jetliner operations for about $2B, and the deal will go through in January if the U.S. and ROC Governments agree. Purchase of such a large and important industry would have a significant impact on Taiwan's industrial base and would certainly require heavy supercomputing usage. In a related matter, a recent announcement in a Taipei paper asked for a bid on supersonic wind tunnel measurement systems. The country already has at least one supercomputer and several dozens of U.S. mini-supers.

Since 1990 Taiwanese companies have invested more than $400M in the Bay area of northern California. Billions more have been invested in Southeast Asia; in 1990 Taiwan became Malaysia's number one investor, $2.3B in that year. More than $500M has been invested in Vietnam since it was opened to outside investors several years ago. In addition, there has been major Taiwan investment in Mexico, anticipating the open border to the United States.

INTERNATIONAL SYMPOSIUM ON ALGORITHMS '91 (ISA'91)

Almost 140 scientists met at the Academia Sinica, in Taipei, Taiwan, 16-18 December 1991 for the Second International Symposium on Algorithms. ISA'91 was organized by the Institute of Information Science, Academia Sinica, and National Tsing Hua University, with cooperation from the Special Interest Group on Algorithms (SIGAL) of the Japan Information Processing Society. The first symposium was held last year (1990) in Japan as SIGAL International Symposium on Algorithms. The participant distribution was approximately as follows.

[blocks in formation]

Because of the wide distribution of this book, I will only make a few general comments about the papers at this symposium.

The goal of the ISA is to promote a forum for Pacific Rim researchers as well as those in other parts of the world to exchange ideas on computing theory. The last two words here are important in the context of this symposium. While "algorithms" has a wide variety of meanings, ISA participants are concerned almost exclusively with discrete algorithms. The papers focused on sorting, permuting, the discrete aspects of computational geometry, combinatorial optimization, graph traversal, etc. Such algorithms have applications in virtually any problem area in which

counting, enumerating, or selecting is significant. A typical one is to design the wiring path in VLSI to minimize space, crossovers, etc. Although a few papers dealt with parallel or distributed algorithms, these were in the minority.

that contributed. There really wasn't a "ringer" in the bunch.

From the Asian side, my own favorite papers were as follows.

• Inagaki (Nagoya, Japan) et al. on constructing shortest watchman routes in a polygon. The watchman route problem deals with finding the shortest route from a point back to itself, so that every point in the polygon can be seen from at least one point along the route.

• Ho (Academia Sinica, Taiwan) on three-dimensional channel routing (for VLSI), which tries to minimize the number of "vias," connections between different levels of the configuration.

Although potential applications abound, ISA papers actually emphasized theoretical aspects. Coming from a mathematics background, I felt right at home with the tone of the papers, although the techniques used were quite different from those I was familiar with. Most papers presented elegant theorems analyzing discrete algorithms, data structures, or generalized graphs and expressed results as "order" or other limiting relations. Here is a typical definition: "We call such an algorithm 'competitive' if its cost performance is at most a constant multiple of the ideal • cost." There was very little evidence of actually using a computer, except perhaps to experiment as a way of suggesting direction, or verify an analysis. Although the subject is different, this is analogous to the style in some theoretical numerical analysis papers. That kind of analysis can ultimately be important, although research has been much more practically oriented.

To succeed in such research it is only necessary to have capable scientists--equipment and other expensive facilities are secondary. Of course, this is one reason that ill-equipped research institutes sometimes concentrate in these areas (not the case for the host institution). Thus it is not surprising that there is an almost seamless flow of research results moving around the world in this field. There appeared to be a great deal of cross fertilization, including students of Western scientists graduating and establishing positions in other countries. It did seem that many of the key directions were set by Western papers. But the level of sophistication in almost all the papers was very high, and I spotted excellent work from just about every country

[ocr errors]

Chan (Hong Kong) et al. on path algorithms for robots minimizing total distance traversed. This was one of a series of very excellent papers on geometry.

Ibaraki (Kyoto, Japan) et al. on a problem with application to the testing of logic circuits by applyinga (limited) selected set of test inputs. Of course, one wants to make correct decisions with minimum testing.

• Chan (Hong Kong) et al. on how to make use of a hypercube computer that has some faulty nodes to run an algorithm requiring a full binary tree. I was impressed last year with related work from these authors (see my earlier report).

For additional details about ISA'91 contact

Prof. R.C.T. Lee
Dean of Academic Affairs
National Tsing Hua University
Hsinchu, Taiwan 30043, ROC
Tel: +886-35-719134 (0)
Fax: +886-35-722713
E-mail: rctlee@nthu.edu.tw

ACADEMIA SINICA

The ISA'91 symposium, described above, was held in the Academic Activity Center of the Taiwan Academia Sinica (Academy of Science). Academia Sinica was founded in 1928 on the mainland and then moved to Taiwan in 1949. The Academia has two basic missions: conducting scientific research and coordinating scientific activities of other government research institutes and universities. The 19 institutes at the Academia are shown in Table 1.

The older institutes were established first on the Chinese mainland, although only 2, Mathematics and History and Philology, were moved to Taiwan intact; for 11 others personnel and equipment were left behind. Several years before an institute is officially established, a preparatory office is first set up (denoted by "pre" in Table 1). At the beginning these offices are mostly planning, but eventually become full fledged research facilities. Note that not all institutes count administrative staff in the figures given in Table 1, so numbers are rough approximations and may not be comparable between institutes. I was told that the total staffing is well over 1,000, although it fluctuates because of part-time employment.

I was fortunate to be able to have a lengthy conversation with

Dr. Ta-you Wu

President, Academia Sinica Taipei, Taiwan, ROC

Dr. Wu, 84, was trained as a physicist with a Ph.D. from Michigan and published his first book, Vibrational Spectra and Structure of Polyatomic Molecules, in 1938. After World War II he headed the Theoretical Physics Section of Canada's National Research Council and later was Chairman of the

Physics Department at the State University of New York at Buffalo. He has had a pivotal rule in the development of science in Taiwan since the mid1950s and has been President of the Academia since 1983. He was mostly responsible for the science development program that sent thousands of Taiwanese abroad to study and eventually to return and staff the country's universities and research institutes. universities and research institutes. Another of his proposals was to limit all academic administrative appointments to a once-renewable 3-year term. A recent public opinion poll showed that he was one of the two most popular men in the country. In 1984 he won lar men in the country. In 1984 he won the Magsaysay Award, roughly the Asian equivalent of the Nobel Prize. In May 1991 he and U.S. President Bush received honorary Ph.D. degrees from Michigan, and Wu has a wonderful photo on his wall commemorating this event.

Table 1.

Physics

Dr. Wu explained to me that in Taiwan responsibility for development of science and technology is divided in three: the Ministry of Defense has its own budget and research agenda and is responsible for technologies related to military needs; the Ministry of Economic Affairs for various research institutes and science and technology related to industrial development; and the National Science Council (NSC) for all academic programs relating to basic, applied, and social science. Academia Sinica is somewhat special, as it belongs directly to the Presidential House (Taiwan's White House). As Academia president, Wu reports directly to Taiwan's President; thus his voice is loudly heard. The current Academia budget is on the order of $30M annually. President Wu spoke strongly about the distinction between basic (fundamental) research and other approaches; it is clear that he favors and has strongly supported the former.

The Various Institutes of Academia Sinica

[blocks in formation]
[blocks in formation]

Nevertheless, all is not perfect. The original intention of the Academia Sinica was that it would serve as consultant to the President on academic related matters. However, I was told that in the past this function was rarely used, and they have become another (large) research institute. Staff can submit research proposals to NSC for funding support just like any university faculty, but they cannot admit students and do not perform any instructional function. Many Taiwan scientists feel that the resources being devoted to the Academia's research program should be re-oriented toward directly assisting the country's industrial infrastructure. In the area of computing, similar views have been expressed by Taiwan-born, U.S.-based scientists, who feel that much more emphasis needs to be placed on practical systems-building experiences and less on highly theoretical paper studies. My own opinion is that it would be healthy to better tie R&D done at the Academia to that supported by other ministries, as these seem more aligned with the country's needs.

The ISA symposium was sponsored by the Institute of Information Science (IIS). There were informal opportunities to visit its building, but no chance to see any other institutes. Computing facilities were comparable to wellequipped Western laboratories: an N-cube, two Iris graphics and about 50 other Unix workstations, and plenty of PCs. There is access to an ETA-10, which is run by the Computing Center. There is a computer vision laboratory with image scanners, camera, image processors to support two-dimensional animation and three-dimensional visualization work, stereo vision, and neural network research. There is a long-term project in natural language understanding for Mandarin Chinese. The N-cube gives some opportunity to perform parallel processing research.

This is related to architecture design, compilers, and parallelizing various constructions within existing languages. There does not appear to be any significant research in numerical methods, either theory or implementation. There is interesting robotics research related to dexterous manipulation, coordinated motion of multiple robot arms, and simulation. As the ISA symposium would suggest, there are about a half dozen people working on the theoretical aspects of discrete (combinatoric) algorithm development. Finally, there are significant efforts in the areas of real-time operating systems, high-speed networking, software methodologies, and a small effort VLSI layout design.

A new eight-story building is being built to be finished before 1994, sufficient to house more than 50 Ph.D. researchers. The institute publishes an English language research journal, Journal of Information Science and Engineering, which contains articles from Taiwanese researchers (not primarily those at IIS). Electronic access has recently been expanded to Internet service, allowing file transfer, Telnet, etc. While Chinese is the daily working language, almost all IIS scientists can function adequately in English. Many have spent substantial time in the West, and a few speak excellent colloquial English. Information about this institute generally, or the journal in particular, can be obtained from the following.

Prof. Lin-Shan Lee

Institute of Information Science
Academia Sinica

Taipei, Taiwan 115, ROC
Tel: +886-2-788-3799
Fax: +886-2-782-4814
E-mail: hsu@iis.sinica.edu.tw

I was told that the Institute of Information Science is typical of those at Academia Sinica. It is essentially run

by the research faculty with very few administrative duties. It is a research institution. There are no teaching duties because there are no students; instead, research is supported by many full-time assistants. Many faculty do, however, offer courses in nearby universities, and there are the usual seminars, conferences, etc. Many IIS staff were heavily occupied with the ISA symposium and there wasn't much other activity while I was there. I did not visit the Institute of Mathematics, but the printed materials that I was given discuss work in rank inequalities for chordal graphs, probability, algebra, and number theory. There does not appear to be any significant activity in numerical analysis. A quarterly Bulletin of the Institute of Mathematics has been published for almost 20 years, and since 1976 another quarterly, Mathmedia, for less esoteric articles.

With nearly 1,000 scientists in so many disciplines, Academia institutes are individually worth serious examination. The preceding comments are meant to provide background and overview material only.

NATIONAL CENTER FOR HIGH PERFORMANCE COMPUTING

In December 1990 I reported on the plans for a national supercomputer center in Taiwan. Almost 1 year later I decided to see how things were coming along. I met with

Prof. San-Cheng Chang
Director, National Center
for High Performance
Computing (NCHC)
P.O. Box 23-115
Taipei, Taiwan, ROC

Tel: +886-2-3630231 x2189
Fax: +886-2-3661571

E-mail: ascchang@hades.nchc.

ntu.edu.tw

Also at the same address is Dr. Kuo-Wei Wu, who was on leave from Cray Research to help with NCHC's planning. He has now resigned from Cray and is a Deputy Director of NCHC. (A year earlier Wu had explained some of Taiwan's goals in the area of supercomputing.) Chang, Wu, and a goodsized administrative staff are currently housed at National Taiwan University, in the city of Taipei. However, a 15,000 m2 building is almost complete.

This is in Hsinchu Science-Based Industrial Park (HSIP), 70 km south of Taipei, the and site of two major national universities, National Tsing Hua (with about 4,500 students and about 450 faculty) and National Chiao-Tung University. Hsinchu's model seems to be Tsukuba Science City in Japan; there are also somewhat similar places in Korea, Singapore, etc. It is usually referred to as Taiwan's "Silicon Valley L"

I did not visit the park this year but had been in Hsinchu a year earlier and wrote about it then. On a site of nearly 1,000 acres there are about 150 factories. An additional 500 acres have been obtained and plans are to house another 150 companies within 5 years. Presently, production values (%) of companies in the park are as follows.

[blocks in formation]

down the park's walls (and eliminating tenants' tax breaks) to integrate its companies better into the society and also setting up similar parks in other parts of the country such as Kaohsiung. Nevertheless, HSIP is a must-stop for anyone interested in industrial development in Taiwan.

Taiwan is certainly not the first country in Asia to build a nationally accessible supercomputer center, but it hopes to benefit from experiences in other places. Planning took place over 4 years and NCHC is organizationally placed under the National Science Council and will be a national laboratory there. It is expected that in late 1992, when the facility finally opens, it will serve both the academic as well as the industrial community. The emphasis is to be on application expertise. NCHC is operated in 5-year phases, the first ending in June 1995. This phase was budgeted with $89M, about half for computer-related hardware and software systems. NCHC's director is chosen by NSC from relevant university faculty. Under him are five divisions.

• Administration • Operation

• Hardware/Software Management • Research/Development

• Education/Promotion

The last two distinguish NCHC from a 1.7 pure cycle-shop and make it more like one of the National Science Foundation (NSF) Centers in the United States. In addition, special interdisciplinary laboratories and projects will be established as needed. Current plans are for a scientific visualization laboratory, and projects in molecular modeling and public domain mathematical library development, and these are to be integrated within and across the divisions. Staff will be about 100, similar in scale to the NSF Centers.

HSIP, like other commercially based high-tech industrial parks with many tenants, has some ailing companies, some with products that are less hightech than authorities might want (one company wanted to develop robots and is now engaged in making windowwashing machines), and some whose high-tech 3 years ago is now just labor saving. There is talk both about tearing

NCHC is to be hooked up to a local network (FMAN) within Hsinchu's industrial park. In addition, it will be hooked to TANET (Taiwan Academic Network), a T1-based public network sponsored by the Ministry of Education and part of Internet. In addition, there will be long-haul T1- or T3-based dedicated lines set up to major universities and research institutes that need such access. NCHC will be internally networked with the Fiber Distribution Data Interface (FDDI).

Chang and Wu emphasized to me that one of NCHC's major functions is to provide expertise. They expect 1,200 potential users during 1992. This will have to be all in the last third of the year, as the hardware is not scheduled to come on line until late summer. By 1995 they are hoping to support at least 2,000 users. Major applications are expected to be molecular modeling, computational fluid dynamics, electron device simulation, structural analysis, atmospheric science, and earthquake wave studies. They are anticipating the usual bevy of academic exchanges, newsletters, seminars, etc., not only focused on how best to deal with existing applications on NCHC's equipment but also anticipating new computing environments such as highly parallel computers that will become user-practical in the future.

At this date a final selection of hardware has not yet been made. The expected vendors have all submitted proposals and these are now being evaluated. The specifications call for a high-performance computing engine, a front-end system, and a large number of workstations for visualizations, computing, and software development. The supercomputing engine will be installed in two stages, 1992 and 1994, to ensure that new equipment will be installed as it becomes available.

« iepriekšējāTurpināt »