Channel, Issue No. 16

Abstract

[evdate 1993-02-15] The Hong Kong University of Science & Technology

Channel Newsletter of the Centre of Computing Services & Telecommunications FEBRUARY 1993 Issue No. 16 Happy Chinese New Year Focusing On: Scientific computing Seasonal Greetings From CCST Clear Water Bay Road, Kowloon, Hong Kong channel@usthk.ust.hk

2 Channel Issue No. 16 In this issue . . . . . 3 Introduction to Scientific Computing 3 Our Stance in Scientific Computing 8 Scientific Plotting Software in HKUST 10 Scientific Computing Application 11 Unix News Unix Servers 11 Staff News CCST Hotline (6200) Mon - Fri : 09:OO - 12:45; 14:00 - 17:00 Sat : 09:oo - 11:45 Computer Barns (Rms 4382 81 1044) Ooenirx Hours Mon - Fri : 08:15 - 22:45 Sat : Ok45 - 16:15 Consultation Hours Mon - Fri : 09:OO - 12:45; 14:00 - 17:00 Sat : 09:oo - 11:45 Business Computing Laboratory (Rm 4334) Opening Hours Mon - Fri : OR15 - 22:45 Sat : OS:45 - 16:15 Editor's Notes Channel is a bi-monthly publication of the Centre of Computing Services and Telecommunications (CCST) of HKUST. Prepared by the Computing Information Centre of CCST and printed by ETC, Channel provides information about CCST plans, developments and services. Topics on new trends in computing technology and other related topics of general interest are also included. User contributions to Channel are welcome and should be sent to the Editor along with the author's name and department. Channel is distributed to University members who are registered users of CCST services. Other parties who would like to have their names added to our mailing list for Channel may complete and return the form on the last page to the Computing Information Centre. Rita Chu, Editor of Channel ClC ccrita@usthk.ust.hk

Issue No. 16 Channel 3 Introduction to Scientific Computing Paul Kwan cckwanwh@usthk.ust.hk Scientific computing has in the past decade drawn much attention from researchers across a broad range of disciplines including those from the physical sciences, mathematics, computer science, and engineering. It was largely an outgrowth of at least two essential factors. First, the widespread use of computers in carrying out numerical experiments that are taking their place alongside the more traditional methods of research. Along with the theoretical and experimental, there is now a computational aspect of science. Second, the steady increase in processing speed of computers brought about by advancements in computer hardware. Since the introduction of UNIVAC I in 195 1, the first commercial electronic computer, the speed of the fastest computers has, on the average, doubled every two years. These, along with other factors, have brought about the evolution of scientific computing into a new academic discipline. Throughout the years, many events were influential to the development of scientific computing including the construction of the early batch and mainframe computers. Almost without exceptions, they were aimed to provide the ability for manipulating numbers at a rate that compresses a lifetime's worth of pencil-and-paper calculations into perhaps a second. The unprecedented speed offered by these computers had enabled scientists and engineers to explore new questions through the use of computer modelling and simulation. In many cases, the results produced were inconceivable decades ago. However, the influence of batch and mainframe computers has diminished with the introduction of lower-cost computers in the last 10 years. There exist today specialized computers known as array processors which can deliver a speed comparable to these early computers when performing large- scale numeric calculations. Graphics workstations joined the scene by combining the abilities for visualization and high-speed computing in a single package. Altogether. the) constitute a strong driving force behind the proliferation of researches in scientific computing as we have witnessed in the past years. Most researches in scientific computing begin with the construction of an elaborate mathematical model -- a set of equations that accounts for whatever is under investigation, whether to figure out the patterns of circulation in the world's ocean or the buildup of carbon dioxide in the atmosphere. This mathematical model is essential as it provides the framework from which a program will be written to simulate the problem efficiently on a computer. Through using a sophisticated graphics program, the numerical results could later be presented in a visual form that the human mind can comprehend. Although these tasks could take months or even years to complete, the insights produced are so useful that they often outweigh the efforts having spent. Harvests of this kind have been widely seen in areas like computational chemistry, physics, astronomy, just to name a few. --

4 Channel Issue No. 16 Among the many progresses made in scientific computing, ) computational chemists can now simulate the behaviour of molecules, atoms, and even sub-atomic particles. Adapting the study of this problem to research in scientific computing begins with a mathematical description of the matter under investigation. This is followed by the process of constructing the sets of equations in quantum mechanics that could describe the structure of each molecule being studied and their interactions. By solving the groups of equations ' numerically, perhaps on a supercomputer, researchers can ~ gain new insights into the architecture of molecules, the 1 course of their interactions, and even predict the properties of compounds that have yet to be produced. In another development. astronomers create mathematical models to describe the interplay of heat, motion, gravity and other factors that underlie the formation of a star, a happening that is largely hidden from an astronomer's view. Through repeated simulations, researchers can trace the birth process of the star beginning with the initial buildup of gravity that I eventually causes an enormous interstellar cloud of dust and / gas to start collapsing in on itself. By adjusting the groups of equations that constitute the model, computerized clouds If various shapes and rotational speeds can be generated. I Experiments of this nature have suggested why some clouds yield gravitationally linked twins called binary stars, while others collapse into solo stars like our own sun.

What you have read in the recent paragraphs represent only two out of many current research efforts in scientific computing. As the number of positive reports continue to increase, we will see growth in several directions. First, the number of researches that could be seen as applications of scientific computing mill increase when more researchers are willing to venture into the new discipline. Second, there will be growth in researches on better numerical algorithms that can both increase the efficiency and impro1.e the accuracy of computer simulations. Third, a growing interest among computer scientists in exploring massively parallel architectures aiming to increase the raw speed of a computer could be anticipated. Fourth, we will certainly be witnessing a tremendous growth in research efforts on computer visualization and its impacts to scientific computing. These are just a few of the many channels through which researches in scientific computing could be directed in the years to come. The scope of such developments would only be confined by the limits of our imagination.

Our Stance in Scientific Computing Danny Tang ccdanny@usthk.ust.hk Author's Note: This article focuses on the generic aspect of scientific computing -- storage, mathematical analysis, and graphical processing of scientific data. It does NOT intend to cover the more specific areas such as finite element analysis, chemistry modular modelling, and DNA sequence matching. If you are a theorist, the industry's recent development in symbolic-based packages may be of interest to you. Scientific computing is not only numerical. All numerical problems begin with a symbolic representation in terms of formulas, equations, and so forth. Some problems are best solved in the analytic domain, many benefit from a mixed approach, with parts of the problem handled by symbolic manipulation, followed by numerical approximation.

Issue No. 16 Channel 5 In the university, we have the two most well-known symbolic packages -- Mathematics and Maple. Apart from their capabilities to solve algebra and calculus equations symbolically, they also allow you to specify the exact degree of accuracy with which numbers are calculated and expressed in solving numerical problems. Both Mathematics and Maple integrate numerical, symbolic, and graphical capabilities with a consistent syntax to assist you in the modeling of real-world problems. However, they are interpretive languages, they may not provide the "number- crunching" speed that you would like to have. If you write your own scientific programs Scientific computing has for many years been done primarily in dialects of Fortran. Many other languages are now also used (e.g. C), but the Fortrans probably still dominate. Both Fortran and C compilers are available on almost all our platforms. One common need for scientific programming is the availability of libraries of subroutines for common numerical algorithms and graphical processing. They are building blocks. Besides the commercial libraries -- MATH, STAT, Exponent Graphics from IMSL, and NAG from NAG Inc., you may find the collection of numerical subroutines described in the book The Numerical Recipes from Cambridge University Press useful for incorporation into your programs. In the future, although subroutine libraries will continue to be important in developing scientific programs, more and more scientific programs can be assembled out of larger units, subsystems and packages. There are packages which are specially tuned to mathematics. Depending on your specific needs, they may be good alternatives to writing programs from scratch using, say, Fortran. An example of such software is MATLAB (for MATrix LABoratory), which is an interactive matrix manipulation package. MATLAB is an useful package since many problems can be elegantly represented using a matrix formulation. On top of MATLAB, many toolboxes have been developed for different scientific/ engineering fields, including signal processing, robust control, and even chemometrics study. Another popular package worth-mentioning is Statistical Analysis System (SAS). It is a comprehensive package for the understanding of experimental data. The software has modules for doing a variety of data analysis such as time series analysis, multivariate analyses and graphical presentation . The software is also available to our university users. What if you only need a handy mathematical/ graphical tool The symbolic package such as Mathematics is powerful, but it requires some effort to harness. If you only need a mathematical scratchpad, then a powerhouse might not be appropriate for you in view of the time you might need to spend in learning it. A good alternative is MathCAD. Although not for complicated mathematical analysis, MathCAD's intuitive interface and wide variety of numeric functions make it an ideal, intelligent mathematical scratchpad. The package also has quite good graphical capabilities and some limited symbolic features. MathCAD runs on MSwindow, and is available to users now, How about graphics support? The trend away from batch processing to interactive working on a window-based workstation has brought graphics display technology to the scientist's desk as the norm. Application of computer graphics in scientific problems has been commonly referred to as scientific data visualization. It is an aid to understanding, so it is part of the exploration

6 Channel Issue No. 16 process, not simply a means of presenting the final results. It is the use of computer graphics technology as a tool for comprehending data obtained by simulation or physical measurement. Some modem visualization packages also allow users to control or steer the simulation and observe the outcome visually at once. Many scientific packages, generic or specialized. 1lav.e graphical capabilities. For example. the software package PATRAN that the Department of Mechanical Engineering procured recently for finite element analyses has a good support for displaying in graphical form the results of finite element studies. More generic packages like Mathematics. Maple, and MathCAD also have an integrated graphical component. On our campus, we have a software product called IMSL/IDL which allows users to do a variety, of plottings. The software also has some image processing and animation capabilities. Please read the next article, Scientific Plotting Software in HKUST, for more details about IMSL/ IDL. How should scientific data be stored/managed This is a tough question. Big scientific problems consume and generate huge amounts of data, much of which must be stored. In a typical program run, a significant portion of the time may be spent in doing data input and output. Huge data resources are also common in business applications, but some types of data (e.g. floating point numbers, formulas and equations, graphs) and the need for rapid access to ordered data sets are particular to scientific computing. The traditional approach of using flat sequential files for data storage is inefficient in storage and access, though it is still the most widely used storage method. The scientific community started to address this issue only recently. One such effort is that of the National Center for Supercomputing Applications (NCSA) in developing the Hierarchical Data Format (HDF) specification. HDF is based on a scientific data model and defines a multi-object file format for transfer of graphical and floating-point data among different systems. It allows, for example, a grid of data to be stored together with scales. annotation, etc: slices through datasets can be extracted. Fortran and C calling interfaces are provided. Also, some useful visualization tools accompany HDF. HDF has been adopted by many universities and research organizations since it defines a framework for continued future growth based on its self-describing capabilities. If you would like to look into this area, you are welcome to discuss your needs with us. Now, it's time to publish your result Congratulations! Scientific word processing is another area that we have looked into. Scientific papers have unusual typographical needs such as representing matrixes and complex fractions, which present tricky placement issues. Scientists frequently use TeX. Many book publishers and technical journals accept TeX files directly. It is an attractive alternative because one does not need to proofread a typeset copy. We provide TeX on the Sun platform. Users of other platforms including

Issue No. 16 7 PC/Mac can run TeX out of our Sun server. A number of style files including the one from the America Mathematical Society are provided. A facility for incorporating postscript codes is also available. Note also that many mathematical packages such as Mathematics and IMSL/IDL can generate TeX output directly. Formulas Marhemarrca can solve problems In algebra and calculus r ~n[!l:= Factor[x^6 - yA6/ 2 2 2 2 cut ill= (x - y) (x + yl (x -xy+y) (x +xy+y) In/Z Integrate[x/(a + Exptxl), xl X 2 x Log[l + FXl PolyLog[2, -+I / Out[21= & - I a a L In[3 Solve[(xA2 + yA2 == 1, X + 2 y == l/r IX, YII 3 4 Out[_l]= {{x -> y I, ix -' 1, Y -' 01) 5 You certainly have our support I have mentioned a number of software packages above, they are all available to you. More information on the software such as the platforms on which the software are available, the command to start particular software and so on, can be found in CCST On-line Information, an information service accessible from any PC or workstation. It is always our objective to develop good support services in a number of application areas, particularly in office applications, programming, mathematical and statistical analysis, graphics and scientific visualization, and data management. Good support means that we have staff with a good general understanding of the technology and the industry's trend, and is able to give users specific advices or pointers for further advices. Users can also expect that support staff exist in CCST with reasonable level of knowledge in operating a few carefully selected software in each of these application areas. In each area, there are hundreds or even thousands of software available from different sources, it is impossible for us to have working experience in even a handful of them. To better focus our support, we plan to develop our expertise and services in the following software initially, and as time goes by gradually take up other packages while resources permit. Package Purpose Mathematica MathCAD SAS IMSL libraries IMSL/IDL TeX Symbolic, numerical & graphical processing Mathematical scratchpad Statistical analysis Scientific programming Graphics & scientific visualization Scientific word processing Your input is always one of the most important elements in helping us determine what software to provide in the university, and how support services should be operated. In the future issues of Channel, I will say more on our plans and progresses in other computing areas.

8 Channel' Issue No. 16 ' Scientific Plotting Software in HKUST A graph is worth a thousand words. As hardware and software are getting more and more advanced, computer graphics has become an integral part of many software systems, like: Animation Business Presentation CADKAMKAE Data Visualization GUI (Graphics User Interface) Image Processing Scientific Graphing Simulation Let me discuss in the following the scientific graph and data plotting capability of some available application software in the university. SigmaPlot SigmaPlot is good for scientific graphing. It produces 2D/ 3D graphs and charts with error bars, log/semi-log and linear scales, polynomial curve fittings, datapoints, and Greek and mathematical symbols. It allows multiple plots per graph and multiple graphs per page. The Data Worksheet, like any spreadsheet, consists of cells, and is over 64,000 rows by 16,000 columns in size. The cell contains the data used to create graphs. One can enter data into the worksheet in a number of ways: - Use the keyboard to type in data directly; - Open a SigmaPlot file saved from a previous session; - Import data from other program files like Lotus, Excel, dBASE and ASCII; - Create data using a mathematical transform Pa&k Lang ccpleung@usthRust.hk Example of a basic transform equation: t=data(-10,11,1.5) ,-generate serial data col(l)=t ,-put t into column 1 f(x)=2*ti3-7 2+9*x-5 ;define the equation col(2)=flt) ;put the result into column 2 The cells will have the following result: 1 21 - 10.00000 -2795.OOOOO 2 -8Soooo -1815.50000 3 -7.00000 - 1097 .OOOOo . . . 13 8.ooooO 643 .ooOOo 14 9.50000 1163.50000 15 11 .ooOOO 1909.00000 The transform sample data is plotted as a curve:

Issue No. 16 Channel 9 Gnuplot Gnuplot is a public domain command-driven interactive function plotting program, capable of two- and three- dimensional functions and data files. It supports many different types of platforms, terminals, plotters and printers, and is easily extensible to include new devices. Among the list of mathematical graphics package introduced here, Gnuplot should score lowest in functionality and GUI, but it supports an exceptional output driver. It can generate graphs in LaTeX format, so users can easily input those picture files into their LaTeX file. The command procedure to generate a LaTeX picture file: set terminal latex set output "examplel.tex" plot [-10:11/2*x"*3-7*x""2+9*=c-5 Mathematica Mathematica is a general computer software system and language intended for mathematical and other applications. It incorporates more than 840 functions. covering virtually every conceivable math operation. One can use Mathematics either by setting up a Mathematics program using variables. loops and conditional statements or interactively by typing a command and getting back an answer. The example graph can be generated by entering: Plot[2*xA3-7*.v'2+9*x-5,/x,-IOJl}] Mathematica represents graphics in terms of a collection of graphics primitives. The primitives are objects like Point, Line and Polygon, that represent elements of a graphical image, as well as directives such as RGBColor, Thickness and SurfaceColor. Each complete piece of graphics in Mathematics is represented as a graphics object. The InputForm command will show the above picture is represented by the Graphics object of a list of Line. Mathematics can produce not only static images, but also animated graphics. The basic idea is to generate a sequence of "frames" which can be displayed in rapid succession. To produce a good sequence of frames for animation, it is important to ensure that the scale is the same in each piece of graphics by setting PlotRange explicitly. The default value is Automatic. IMSL/IDL IMSL/lDL is an integration of IMSL's C/Math/Library with Research System's IDL (Interactive Data Language) visualization system. It allows users to view data in a variety of graphic formats ranging from simple line plots to complex contour and surface plots and images. It offers a variety of image-processing and signal-processing routines, including contrast enhancement, smoothing, histogram equalization, convultion filters, erosion and dilation operators, and geometric operations that magnify, rotate and warp images. lMSL/IDL allows the construction and manipulation of simple graphical user interfaces. It has a graphical widget set with sliders, buttons, labels, pull-down menus and drawable widgets that sene as the basic building blocks for creating windows. This style of interaction offers many significant advantages for "what-if" experiments.

10 Channel Issue No. 16 Scientific Computing Application Technology has made sophisticated missions possible, things that are impossible a decade or two ago has now become possible. The computer technology has progressed so rapidly that hardware cost is inexpensive such that very powerful computers like super-computers have become affordable not just for military or big research projects. Consequently, scientific computing will be the leading trend for the years to come. Scientific computing has a wide range of applications, among these, real time systems are one of the most common usage. Real time systems such as process control switching, robot arms control. image analysis, air traffic control and so on are gradually replacing the manual operations. In the Research Centre of the Research & Development Branch, there will be a couple of projects which require scientific computing technology. It has been anticipated that there will be collaboration with NASA to set up a satellite station in UST for receiving high resolution image data from the Ocean Colour Satellites. This station will eventually be developed also to receive data from the Earth Observation Satellite for the purpose of environmental study. There also have been discussions on the weather warning system for the new airport. These projects will involve extensive real time technology which has the following features and characteristics. Real time computer systems are very different from other computer applications such as telephone billing, inventory control which do not depend on the time scale of the input in relation to all the work that must be done by the system. In order to handle sporadic, spontaneous, asynchronous events, a real time system must be able to interrupt the current activities immediately and respond to the input because any delay could cause catastrophe. Therefore the environment should be able to manipulate numerous tasks simultaneously such that each task has its own data area, stack, program counter and so forth, and yet the task synchronization should be invisible to the task. Rapid response to external stimuli is another inherent feature of the real time systems. The tolerable time to recognize an external interrupt and suspend the current task should be in the magnitude of microseconds. In many cases, the average processing time exceed the capability of the processor, Ophelia Lee ccophe@usthk.ust.hk therefore it employs multiprocessors working in unison to share the application load. The benefit of multiprocessors is the significant improvement in the overall system throughput, however, the improvement in performance is never as great as the number of processors. Some multiprocessing environment is designed in such a way that the numerous processors work as if they are one powerful processor by connecting to a common backplane. While other multiprocessing environment will be set up such that one of the CPU's will be the master which will assign work for the remaining CPU's to act as slaves. In a multiprocessing environment, tasks can be made global or local. Global tasks are created to run on any processor while local tasks can only run a specific processor because a local task has its own local stack and local variables. As a result, local tasks can reduce the backplane contention which will increase the throughput. On the other hand. global tasks will facilitate automatic load levelling which also increase the throughput. So whether a task should be made global or local depends on the nature and function of the task. In most real time applications, a characteristic which cannot be ignored is the massive volume of data which must be captured and processed instantenously. Hence, a file management system is required to handle a generic mass storage device. In order to optimize the disk writing process of storing the data, the storage volume should have ample of contiguous space to minimize the track searching time. Subsequently, data are stored in an unstructured way and this will require a file management system which implements a set of storage rules and conventions set by the file system. There are many industrial standards and proprietary file systems that are used in real time applications. Last but not least, reliability is one of the most important criteria in real time systems because they cannot allow any system failure due to hardware or software problems as any failure could be disastrous. Hence when designing real time systems, the back up system which can either be another identical system or can even be a manual system, must be clearly defined and well tested. It would be meaningless to have a backup system which will not work during a system failure.

Issue No. 16 Channel 11 Unix News Unix Servers To improve Unix services, 3 new SUN servers were added to the line of servers recently. They are: + ustsu5 (uststu) 0:. ustsu 7 + ustsu8 (stu) There is also a functionality reshuffle in the old Unix servers. After all these changes, the staff and postgraduate mail server is separated from the undergraduate one, thus improving server robustness and stability, as well as email response time. The whole picture of the Unix servers now looks: ustuxl 33 MIPS DECstation 5000/240 As undergraduate Pathworks mail server, Pathworks file service and printer server. ustsul (uxmil) 28 MIPS sparc2 As staff and postgraduate mail server (previously staff, postgraduate and undergraduate mail server), AFS database and file server, news server. Internet name server for UST. and staff and postgraduate file server. ustsu2 28 MIPS sparc2 As undergraduate file server. ustsu3 (uststyg 28 MIPS spare2 As dedicated staff and postgraduate time- sharing machine with undergraduate logon disabled to give a stable environment. ustsu4 28 MIPS SUN IPX As undergraduate time-sharing machine. ustsu5 (uststu) 86 MIPS spar-c IO As undergraduate time-sharing machine to raise performance. ustsu 7 28 MIPS SUN IPX As dedicated NIS server to improve Unix system robustness and provide faster response time for most user activities involving network information access. ustsu8 (stu) 28 MIPS SUN IPX As undergraduate mail server. All time-sharing machines are now upgraded to SUNOS 4.1.3 implemented with a lot of OS bug fixes on 4.1.1. Staff and postgraduate students can log on ustsu4 and ustsu5, but they are not allowed to run email there. With the split of the staff&postgraduate/undergraduate mail servers, staff members and postgraduate students should now use username @ uxmail. ust. hk as their mail address, and, undergraduates should use username@stu.ust.hk as theirs. Staff News Mr Raymond Chau, Computer Officer, married Miss Alice Chic on 5 December, 1992. Mr Andrew Tsang, Senior Computer Officer, announces the birth of his baby boy, Conan, on 11 December, 1992.

12 Channel Issue No. 16 CCST Contact Points CCST Contact Points User Consultation Coordinator - Mr. Tony Chan - 6243 Telephone Services Coordinator - Ms. Christine Cheng - 6190 E-mail cctony cccheng Problem Reporting E-mail Account - CCHELP For users to send in their problems and difficulties via e-mail. A consultant will respond to messages sent to this account as soon as possible. Suggestion E-mail Account - Training Courses Registration - 'CHANNEL - Ms. Anna Mak - 6189 ccanna For users to send in their queries, ideas, suggestions

and comments concerning services we provide. A consultant will respond to messages Sent to this account Requests for Computing Resources - (Admin & Business) Mr. William Tung - 6221 (Others) Mr. Danny Tang - 6241 ccbtung ccdanny as soon as possible. 2 Phone

E-mail Centre of Computing Services and Telecommunications. . . . I E-mail 1 Phone address Director Dr. Wm. Max Ivey 6182 ccmax Assoc. Director & Mgr. Systems & Operations Mr. Lawrence Law 6201 cclaw Mgr. Information systems Mr. William Tung 6221 ccbtung Mgr. CIC Mr. Danny Tang 6241 ccdanny Mgr. Systems Engineering Mr. Michael Tang 6261 ccwctang Editor of Channel Miss Esther Chan 6242 ccesther FAX 358 0967 Dialup 358 2440 Computer Operation 6220 SE workshop 6280 The telephone extensions are (852) 358-xxxx. The Internet addresses are E-mail aaUress@usthk.ust.hk Channel Mailing List A mailing list is maintained for the distribution of Channel. To be placed on the mailing list*, fill out this form completely and mail to the Computing Information Centre, CCST, Hong Kong University of Science and Technology, Clear Water Bay Road, Hong Kong. Please print clearly. 0 Add my address to the mailing list 0 Address change (write new address below) q Remove my name from the mailing list Name : Organization : Address : Registered users of the CCST services are placed automatically on the mailing list.