« iepriekšējāTurpināt »
I am very optimistic about the future of this center. The planners have emphasized that they are NOT interested in building Taiwan's own supercomputer or engaging in advanced computer science software research, but instead want to concentrate on the research associated with applications. NCHC staff are already "on the road," providing training via courses and seminars on techniques related to high-performance computing. They are funding university faculty to port and tune their existing programs to make better use of whatever supercomputer is going to be installed. (Typically, existing programs will not take maximum advantage of vector or parallel architectures.) I am optimistic also because in recruiting staff for the new center, double talents are being emphasized, application fields first and then good knowledge about computer architecture, algorithms, etc. The hope is to get advanced computing to the end user quickly. User groups based on application area are being formed. Finally, the vendor that is finally selected will be required to provide not only systems support but application research support, researcher exchange, seminars, and so forth.
David K. Kahaner joined the staff of the Office of Naval Research Asian Office as a specialist in scientific computing in November 1989. He obtained his Ph.D. in applied mathematics from Stevens Institute of Technology in 1968 From 1978 until 1989 Dr. Kahaner was a group leader in the Center for Computing and Applied Mathematics at the National Institute of Standards and Technology, formerly the National Bureau of Standards. He was responsible for scientific software development on both large and small computers. From 1968 until 1979 he was in the Computing Division at Los Alamos National Laboratory. Dr. Kahaner is the author of two books and more than 50 research papers. He also edits a column on scientific applications of computers for the Society of Industrial and Applied Mathematics. His major research interests are in the development of algorithms and associated software. His programs for solution of differential equations, evaluation of integrals, random numbers, and others are used worldwide in many scientific computing laboratories. Dr. Kahaner's electronic mail address is: firstname.lastname@example.org
JAPAN ATOMIC ENERGY
Computer-related research at the Japan Atomic Energy
Research Institute (JAERI) is summarized.
by David K. Kahaner
JAERI's budget and staffing grew Tokai-mura, Naka-gun, SUMMARY
rapidly through the mid-1980s but have Ibaraki-ken 319-11, Japan
essentially been constant since 1985, Tel: +81-292-82-5611 The Japan Atomic Energy Research nearly $880M (about 90% from STA) Fax: +81-292-82-6070 Institute (JAERI) was established in and about 2,500 staff (one-third each E-mail: email@example.com 1956 to perform the research and devel- research, technician, and administraopment (R&D) associated with imple- tive). JAERI is currently building a I also had an opportunity to meet with menting the country's nuclear program. high temperature test reactor, operat- three members of his research staff, It is funded under the Science and ing fusion experimental equipment, Technology Agency (STA), which is an building a synchrotron orbital radia- Dr. Mitsuo Yokokawa arm of the Prime Minister's Office. tion (SOR) facility, etc.; plant and Dr. Hideo Kaburaki
Currently, the major projects are as equipment are expensive and account Mr. Hiroo Harada follows. for the very large budget.
(same address as above)
Japan generates approximately 30% Tel: 0292-82-5976 • Nuclear energy production system of its electrical power by nuclear means,
Fax: 0292-82-6070 including high temperature gas- although this figure would reduce to E-mail: yokokawa @catalyst.tokai. cooled reactor and fusion reactor 17% if other electrical generating plants
jaeri.go.jp were used to capacity. Sentiment in the • Nuclear safety
country is modestly antinuclear, but and also with
the major reason that JAERI budgets • Radiation applications
have not increased is that nuclear tech- Dr. Masashi lizumi
nology is now mature. The nuclear ship, Deputy Director General • Nuclear ship
MUTSU, has been behind schedule Tokai Research Establishment
and immersed in controversy for years; Japan Atomic Energy Research JAERI has an administrative head- it is now scheduled to be decommis
Institute quarters and a radioisotope school in sioned in the spring of 1992.
Tokai-mura, Naka-gun, Tokyo. The main research center and a I spent 1 day at Tokai visiting with Ibaraki-ken 319-11, Japan fusion research center are in Tokai and scientists in the Computing and Infor- Tel: +81-292-82-5014 Naka, small towns on Japan's Pacific mation Systems Center (CISC), which Fax: +81-292-82-6111 coast about 70 miles north of Tokyo. is essentially the computer support Radiation chemistry research occurs organization. They also perform some Interestingly, Dr. Iizumi and I had some both near Osaka and at Takasaki, half- of their own research. CISC is run by common background, as he has visited way between Tokai and the Japan Sea
the reactor facility at the National Instiside of the main (Honshu) island. Dr. Masayuki Akimoto
tute of Standards and Technology Nuclear ship research occurs at the
(NIST) in Maryland several times. northern tip of Honshu. (Japan's nuclear Computing and Information
During the time of my visit, Prof. waste disposal facility is also near the Systems Center
Eugene Wachspress (University of northern end of Honshu, but this is not Japan Atomic Energy Research Tennessee) was also visiting JAERI in run by JAERI.)
order to give a series of lectures on various topics in the area of numerical (56 Kbps) for several years and is cur- intelligent nuclear power plants; and linear algebra related to reactor prob- rently connecting itself to Internet. There peripherally to create and transfer artilems. As is well known, the nuclear are also two DECnet links, one serving ficial intelligence (AI) techniques in industry provided some of the earliest the fusion research community (I the nuclear field. Akimoto pointed out problems in numerical analysis and commented in an earlier report that that a paper on HASP has been subwas a significant motivator for the devel- worldwide this group is heavily DEC mitted to “Expert Systems and Comopment of large-scale supercomputers. and VAX oriented), another that sup- puter Simulation in Energy EngineerThere are still major problems associated ports TCP/IP. (Interestingly, JAERI's ing,” 17-21 March 1992 at Erlangen, with all aspects of nuclear energy that earliest computer was an IBM 650, the Germany. require extensive computer analysis. same model that I began computing on Computer integration of nuclear While new civilian power plants are at the Watson Laboratory, then at design programs. Building a nuclear not being built, there are many Columbia University.)
reactor is a complex process involving applications of nuclear energy. Also CISC has a total staffofabout 55, 10 engineering from diverse areas. Curthere are many research reactors and involved in R&D (including 2 senior rently, JAERI scientists are constantly medical applications. Finally, there are scientists) and 20 in systems, networks, running large programs, taking the the knotty questions about disposal, etc. The remaining 25 are contractors, resulting output ofone and using parts leakage, etc.
mostly for operations. Some interest- of it for yet another calculation. One In his lectures, Wachspress focused ing statistics for last fiscal year follow. CISC project is to integrate the proon sparse matrix methods for linear
grams (which are frequently produced equations and also for eigenvalue No. of users
1,855 by different commercial vendors) and problems. These techniques are often CPU time used (h)
use expert systems to ease the transiat the heart of nuclear application codes. No. of time sharing sessions 526,619 tion between different programs. Another purpose of his visit was to No. of batch jobs
659,265 Vectorizing. Much of the work at assist JAERI in thinking about how
CISC centers around getting the best best to use new computers. He remarked CPU Utilization (%):
performance from existing computers to me that there is a great deal of time
on standard workhorse programs (hence and money spent getting old programs Nuclear fusion research 45.3 Wachspress' remark above). Since the operational on new machines. This Fundamental research
19.7 mid-1970s more than 50 nuclear proincludes fitting old algorithms into Nuclear safety research
17.9 grams have been vectorized. To assist parallel and vector architectures when Other
17.1 them in assessing what might be expected new algorithms especially designed for
from the Fujitsu VP machines, CISC these computers would be much more Dr. Akimoto explained to me that has assembled a suite of benchmark efficient. This is a common problem at most of the applications software work programs for testing vectorization. The any laboratory that has a significant is still being done in Fortran, although list below includes programs of direct development effort in place using new projects within CISC are using C interest to JAERI and also some large, existing program packages. and Lisp. As expected, younger scien- standard benchmark problems that CISC has two Fujitsu VP2600 super- tists prefer the newer languages.
readers will recognize. computers, used mostly for running Research at CISC is centered around large nuclear codes. One is at Tokai, the following.
SLWALF Simulation Code for Alphathe other at nearby Naka where the Robots at work inside nuclear power
Particle Heating fusion research is located. There are plants. (Parts of this environment must AEOLUS Three-Dimensional (3D) Magalso three other Fujitsu M780 series be among the most hostile imaginable.)
netohydrodynamic Code mainframes and many workstations. In Working in Lisp, CISC scientists have STREAM Thermal Hydraulics Analysis Tokai the computers are served by a been simulating how human-acts could
Code 157.5-GB disk, 88-GB optical disk, tapes, be realized by a human-shaped intelli- VDIRECT High Temperature Test Reacetc., and there is similar equipment at gent robot (HASP: Human Acts Simu
tor Code Naka. There is a 6-km on-site local lation Program). This is part of a 10-year SRAC Thermal Reactor Standard area network (LAN) (205 Mbps) and a program begun in 1987 to develop basic
Neutronics Code System 9.2-km optical fiber connecting Tokai and underlying design technologies for CITATION 3D Neutron Diffusion Code and the fusion laboratory. The Tokai intelligent robots; to develop the basic TWOTRAN Two-Dimensional (2D) Neulaboratory has been connected to Bitnet technologies for very advanced,
tron Transport Code
DOT35 2D Discrete Ordinates Neutron
Table 1. Vectorization Ratios and Speedups for
VP2600 and VP100 Computers
Mode BERMUDA Neutron-Gamma-Ray Trans
Mode port Code System
5.0 ISOFLOW Melting Process Analysis Code
3.4 DSMC Gas Kinetics Code by Direct STREAM
4.8 Simulation by Monte Carlo
4.7 SPIN Spin-Spin Interaction in 2D
4.0 Ising Model
5.3 VIENUS Visco Elastic Stress Analysis FPGS
6 SONATINA BERMUDA
3.0 Seismic Response Analysis for
5.0 High Temperature Gas Reactor
3.9 RELAPS Light Water Reactor (LWR) VIENUS
3.3 Transient Analysis Code
4.0 RELAP5 WIND
3.2 3D Wind Field Calculation
7.0 JPEC JAERI Performance Evalua
4.4 tion Code for Basic Operations LLLC
Lawrence Livermore National
Two CISC scientists I met formance Analysis Code Speedup Programs (Yokokawa and Kaburaki) have experi
ence with parallel machines as well Using these programs, JAERI
as traditional supercomputers. Dr. computed the fraction of computation
Yokokawa is a Ph.D. graduate of that was vectorizable and the speedups
Tsukuba University and worked under running on the VP2600 compared to
Professor Y. Oyanagi (now at the their older VP100, both in vector and
University of Tokyo) and Professor scalar mode (see Table 1).
M. Mori (University of Tokyo), two On 40 programs of specific impor
well-known Japanese numerical tance to JAERI, they gave me the fol
analysts. Dr. Kaburaki was coauthor of lowing statistics.
a paper on the use of Fujitsu's AP1000 10-11
parallel processor for Monte Carlo Vectorization No. of
simulation of gas dynamics. This paper Ratio (2) Programs
was presented at the recent joint Fujitsu13-14
Australian National University work50-60
shop. Yokokawa and Kaburaki gave 60-70
me another paper coauthored by JAERI 70-80
and Fujitsu scientists on parallelizing 80-90 2
the Monte Carlo program MCACE for >90
23 Thus most programs result in only the AP1000. This had also been pre
modest speedups. This is fairly typical sented at the aforementioned workAnd the following vectorization speeds for programs that were originally writ- shop. At the moment, Kaburaki is also ups on the VP2600. ten for older scalar machines.
working with N. Ito, another recent graduate from the University of Tokyo's physics department on the development of a special purpose parallel computer (m-TIS) for doing Ising spin computations. Ito's primary thesis work was on using a supercomputer for similar calculations. He was not at JAERI during my visit, so I will report on this work later.
I also had an opportunity to learn about JAERI's scientific subroutine library, JSSL, from Mr. Harada. This was of special interest to me as I have done research in development of similar packages. I was mildly disappointed as this library is not really a current product; its last revision was in 1982. There is only a very small staff to manage and work on the library, and it is a big job to do so. JSSL does contain some excellent programs, such as MA21A originally from the Harwell library, but also some very old routines, including Crout reduction of matrices and Romberg quadrature. It did not appear to have any of the modern, standard programs such as those from Linpack, Eispack, Quadpack, etc. In fact, some JAERI scientists were surprised to hear that in the West, large quantities of high quality mathematical software are in the public domain and can be obtained easily and rapidly by electronic mail. Of course, some scientists are using this approach for their own personal research, but it is far from common, and in any case the routines have not made their way into JSSL. These observations were consistent with comments made to me by Wachspress, who noticed that among the Japanese scientists he met, awareness of developments in the West was uneven. Essentially all open research from any country is known here--and at a high level of absorption--but its distribution appears to be less uniform than in the United States or European Community. Electronic communication will go a long way, in my opinion, to assist scientists in accessing modern information from anywhere in the world. This is a very important trend.