Table 1. shows the changes in budget and staffing levels
from fiscal '91 to '94.
-------------------------------------------------------------------------------
Project Fiscal year Budget ICOT Subcontracted
name Apr. to Mar. billion yen members company members
-------------------------------------------------------------------------------
FGCS(10th year) '91 7.2 90 500
FGCS(11th year) '92 3.6 60('92/10) 200
Follow-on(1st year) '93 1.3 40('93/ 4) 50
Follow-on(1nd year) '94 1.3 40 50
-------------------------------------------------------------------------------
In fiscal '92, the number of ICOT researchers decreased from
90 to 40 and the 7 research sections were reduced to 2 research
departments. In affiliated companies and software houses, the number
of researchers, engineers and managers decreased from 500 to 50.
These 50 people were brought to work at the ICOT research center.
While many researchers and engineers left the FGCS Project, about
30 researchers who had been involved in the FGCS Project at ICOT or in
their companies moved to major universities between fiscal '91 and
'93. Many of them received Ph.D's for their research work with the
FGCS Project. These researchers are playing a very important role in
the Follow-on Project.
2.2 Selection of research themes
As the budget and the number of ICOT researchers shrank, we
had to curtail research themes. In the final stage of the FGCS
Project, we had more than 50 research themes, including about 20
parallel application systems. These included practical research
themes such as expert systems for job scheduling and VLSI CAD systems,
as well as basic research.
The general technical goal of the Follow-on Project is to make
major software systems developed in the FGCS Project operational on
Unix-based parallel machines. Thus, the primary technical goal is the
development of a new KL1 and PIMOS environment named KLIC on
Unix-based machines.
In choosing other themes, selection criteria included whether the
theme would have a large impact on the future of computer science, and
whether progress could be effectively accelerated by the use of
parallel processing.
We chose the following research themes for the Follow-on Project.
They are divided into two groups.
- Parallel Basic Software
- KLIC system: a KL1 programming environment for
sequential and parallel Unix-based machines
- Evaluation of PIM architectures and their KL1 language processors
- Parallel nested relational DBMS, Kappa
- Knowledge Processing Software
- Parallel theorem prover, MGTP
- Knowledge representation languages;
- Deductive and object-oriented language,
- Parallel constraint logic programming language, GDCC
- Heterogeneous distributed problem solving system, Helios
- Genetic information processing systems
- DNA and protein sequence alignment and editing system
- New algorithms for sequence and structure analysis
- Biological DBMS and KBMS
- Legal reasoning system, new HELIC-II
2.3 Research activity of each theme
2.3.1 KLIC system
The KLIC system consists of a KL1 compiler and a run-time
library. The KL1 compiler is written in KL1 and compiles a KL1
program into a C program. The run-time library is prepared as a
library of C programs which provide functions such as debugging,
monitoring, parallel execution management, resource management and so
on. These functions are almost the same as the ones which PIMOS
provides on the PIMs.
Some of these C programs are linked to a user program, if it uses
some of these functions. If the user program does not use any of these
functions, it is compiled into a very simple C program which is
executable even on a small personal computer. This nature of KLIC is
very convenient for educational purposes.
The development of KLIC was done in two steps. In the first step,
a KLIC sequential version was developed. Development of the KL1
compiler was a main topic. This version could attain a fairly good
performance. It attained 2 MLIPS on an SS-10/30 and 3.7 MLIPS on DEC
AXP. The KLIC sequential version was released in November 1993 as
ICOT Free Software.
In the second step, a KLIC parallel version was developed. This
version is more complicated than the sequential version. It uses the
software called PVM which provide us with a standard interface for
inter-processor communications over parallel or distributed machines.
This software is distributed as PDS and is provided for most of the
recent parallel machines.
A new KLIC run-time library for the parallel version was developed
to provide functions for parallel execution management, resource
management, debugging and monitoring, and so on. First release of the
KLIC parallel version was made in September 1994 and will be released
in February 1995 as ICOT Free Software.
The KLIC parallel version will be ported to parallel machines such
as Sparc Center, DEC AXP 7000, CM-5, AP-1000+, Cenju-3, SR-2001, SP-2
and so on.
2.3.2 Evaluation of PIM architectures
This theme intended to further investigate five models of
PIMs. All models support KL1; however, the architecture of the
processing element and inter-processor connection mechanism in each
model are different.
Designers of the five PIM models formed a task group and had
interesting discussions. They gathered evaluation data and compared
unique design features with each other. The general evaluation
results are presented at the FGCS'94 symposium.
2.3.3 Parallel nested relational DBMS, Kappa
Kappa is an extended relational DBMS which permits us to use
nested tables to represent data.
In a standard relational DBMS, we usually use two dimensional
tables to represent data structures. One table is divided into many
rectangular boxes by rows and columns. The structure of this table is
very regular and rigid. We are permitted to put only one data item in
each box. This regularity is an advantage as long as we use only
simple data structures.
A nested relational DBMS permits us to put multiple data items in
one box. Thus, it is beneficial to handle complex data structures such
as natural language dictionaries, biological data, and so on. We can
represent complex data in more compact and comprehensive format in a
nested relational DBMS than in a standard relational DBMS.
We developed Kappa on the PIM and made full use of parallel
processing to gain better performance.
In the Follow-on Project, Kappa was almost reconstructed to be a
compact and faster system. Some of its low-level functions were
re-written in C code for better performance.
Now it can attain an almost comparable performance to ordinary
relational DBMSs for a complex database such as GenBank on a
sequential machine. If it is used on a parallel machine, it will
attain even better performance using parallel processing.
2.3.4 Parallel theorem prover, MGTP
MGTP is a model generation theorem prover for full first-order
logic. It is one of the most successful application programs in the
FGCS Project. It demonstrated the effectiveness of the KL1 and PIMOS
environment on the PIMs.
Generally, it has been well known that theorem provers have a very
large search space and thus would be an interesting application of
parallel processing. However, its computational structure is very
irregular and it is hard to predict how its search tree will extend
its branches.
Thus, we need to divide the computation into parallel-processable
processes and their allocate them to many element processors
dynamically on the fly. It is very hard job to do this with
conventional methods.
MGTP successfully implemented this job division and allocation in
the KL1 and PIMOS environment on the PIM model in late 1991. The
execution speed increased almost proportional to the number of element
processors.
Furthermore, the program production period was surprisingly short.
This was the first clear evidence that the KL1 and PIMOS environment
on PIMs can successfully be applied for solving a large-scale AI
problem.
The experiences and skills of the MGTP group were quickly
transferred to other research groups at ICOT. This encouraged all the
researchers at ICOT and their companies.
In the Follow-on Project, the MGTP group further developed their
provers. As a result, MGTP became the fastest theorem prover of this
kind in the world and proved some open problems in quasi group theory.
Furthermore, some of the MGTP provers were provided as tools for
practical applications such as the rule-based engine of the HELIC-II
legal reasoning system.
This indicates that a theorem prover can be regarded as a
higher-level inference engine and that it can be adapted for knowledge
processing applications such as KBMS, natural language understanding,
and software engineering.
In addition to the ordinary MGTP provers, the MGTP group for
educational purposes is developing compact versions of their provers
using Prolog.
2.3.5 Knowledge representation languages
In the FGCS Project, the research on knowledge representation
languages at ICOT has been based on mathematical logic. The research
started from nested relational databases and deductive languages. In
the intermediate stage, research on constrained logic programming
languages was started.
On the other hand, research on object-oriented languages (O-O
languages) at ICOT was mainly done with system description languages
such as ESP, which is extended Prolog developed as a programming
language for PSI machines.
Outside of ICOT, some database researchers were attracted by O-O
languages and their mechanisms for modularization and inheritance.
They developed O-O databases and indicated that they are able to
provide us with more flexible data models than relational databases
and deductive databases. We started discussions on how to introduce
the merits of O-O databases into our framework.
In the final stage of the FGCS Project, we started the design of a
deductive and O-O database language and DBMS. Finally, the design of
this language and system was completed and became the knowledge
representation language, Quixote. The first version was implemented
in KL1.
As Quixote has rich O-O based functions combined with a deductive
language, it can fulfill requirements for describing complex knowledge
fragments such as legal rules and biological reactions. However,
because of the wealth of its functions, implementation was so
complicated that performance and the stability of the software system
were not satisfactory.
In the Follow-on Project, intensive efforts were made to improve
language the specification and system implementation, and the system
was ported to Unix-based machines using KLIC. Quixote will be released
as a more practical system in March 1995.
This system is now called the "big"-Quixote.
Micro-Quixote, which is a compact subset was also developed for
educational purposes. Micro-Quixote is small enough to run on a
personal computer.
2.3.6 Genetic information processing systems
Research on genetic information processing was started with
the parallel processing of multiple alignment of protein sequences.
This research topic is continued in the Follow-on Project and has now
been extended to a sequence alignment and editing system.
This system can handle both protein and DNA sequences. Its
alignment algorithm is based on a DP matching algorithm and
implemented in KL1. Parallel implementation of this algorithm has
been improved many times to attain better performance on a PIM.
Recently, the use of a genetic algorithm has given us better
alignments for some interesting cases. The use of constraints between
some amino acids or nucleic acids was tried in order to narrow the
search space. This system is now being ported to Unix-based parallel
machines and is available to biologists.
Research on the prediction of protein structures has been made in
the Follow-on project. The use of the Hidden Markov Model (HMM) has
given us interesting results in some cases.
Research on biological DBMS and KBMS has also been made in
connection with research on knowledge representation languages. Some
biological reactions as well as the characteristics of biological and
chemical materials are written in Quixote and stored in the knowledge
base.
Reactions were coded into hundreds of figures by biologists. They
are now described in Quixote and can be retrieved by the KBMS. We
expect that this kind of use of knowledge representation languages
will be beneficial in many other application areas.
In the research on genetic information processing, research
collaboration is essential. We have set up several international and
national collaborative research projects with biologists and computer
specialists. They have brou to us many fascinating problems in recent
molecular biology. We hope that the software tools we produced will
continue to assist them with their research.
2.3.7 Legal reasoning system, new Helic-II
Research on the Helic-II legal reasoning system was started in
the final stage of the FGCS Project. It is one of the application
systems developed to generally evaluate the KL1 and PIMOS environment,
knowledge representation languages and other software tools.
This sytem was very successful not only in demonstrating the
usefullness of the FGCS technology, but also in showing us possible
uses for the FGCS technology which are beyond our initial
expectations. This system gives us a better understanding of how FGCS
technology can be used for applications in social-scientific areas.
To analyse a given case and predict all the possible judgements,
Helic-II used two knowledge bases and two inference engines. One is a
case knowledge base combined with a case-based inference engine. The
other is a rule knowledge base combined with a rule-based engine. The
rule-based engine was built using the MGTP theorem prover as its
kernel.
To describe the penal code and case rules, some of the results of
the research into knowledge representation languges were used. Both
inference engines were implemented in KL1 and were speeded up by
parallel processing on a PIM.
As Helic-II used many other research results developed in the FGCS
Project effectively, it was the most appropriate program for general
evaluation of the FGCS technology.
In the Follow-on Project, this research is progressing to include
more sophisticated functions such as simulating a debate between a
prosecutor and lawyer. This extended version is called the new
Helic-II system and requires descriptions of more cases and rules.
To do this, more background knowledge on the law and legal systems
is vital. Thus, collaboration with researchers and experts in a low
area is indespensable. Setting up this collaboration needed more
effort than for genetic information processing because of the wider
culture gap between computer science and law.
The ultimate goal of this research is still very far away, but the
spin-offs from reseach are expected to be used in many social
scientific applications where the problems to be solved will be
simpler than this system. The new Helic-II is being ported to
Unix-based machines and released as ICOT Free Software.
3.1 Framework for dissemination
Around the end of the FGCS Project, MITI organized a
high-level committee to assess the research results of the FGCS
Project. [DPBCT 1994] One of its conclusions was that the results were
considered to be still so far away from the market's needs that
computer companies could not commercialize them in a few years
although they can be highly evaluated from the academic viewpoints.
Thus, dissemination activities were mainly directed at academia,
that is, national and foreign research institutes and universities.
However, quite recently, computer companies both of home and abroad
have become more positive and supportive of our dissemination
activities.
We prepared for dissemination in two aspects. For the first one,
we prepared hardware and facilities such as file servers connected to
the Internet accompanied by a team to maintain the programs and
documentation of ICOT Free Software, and high-speed network links so
that our research partners can use the PIMs from remote sites.
For the second one, we have put our efforts into networking among
ICOT members. We have been able to keep some of the collaborative
research projects which were set up in the FGCS Project and have tried
to set up new ones with national and foreign institutes and
universities based on the new research themes of the Follow-on
Project.
3.2 Collaboration on research in Japan
3.2.1 Organization of Task Groups
In the FGCS Project, we had several "Working Groups
(WGs)" which recruited researchers from academia and industry and
provided them with the opportunity to exchange technical information
and provided some research tools at ICOT.
The main activity of the WGs was to meet and discuss specific
research topics and so the relationship between the WG members and
ICOT was not so tight.
In the Follow-on Project, we needed a tighter relationship with
some researchers in universities and research institutes in order to
to ask them to carry out parts of our research and development or to
help us hold seminars and workshops and so on.
Thus, we set up new Task Groups (TGs) to recruit researchers from
universities and industry. Naturally, many of them had been involved
in the FGCS Project at ICOT or in their companies.
They are playing a very important role in the dissemination
process. For example, we have held KLIC seminars several times at
universities with the help of these TG members.
We currently have the following 7 task groups. The numbers shown
in parentheses are the regular members excluding ICOT researchers.
- Parallel Symbolic Processing System Task Group (18)
- Parallel Inference Machine Evaluation Task Group (10)
- KLIC Task Group (9)
- Parallel Theorem Proving Task Group (9)
- Heterogeneous Knowledge-Base Task Group (17)
- Protein Structure Prediction Task Group (17)
- Legal Reasoning System Task Group (7)
3.2.2 Collaboration with Japanese universities
and research institutes
Some of the research results of the FGCS Project are practical
but some others are not. KLIC is an example of a practical one.
Results are integrated into software systems which can be used as
useful tools. Thus, technology transfer is rather easy.
However, some others were not well integrated into complete
software systems. For example, parallel programming methodology using
KL1 is partially described in technical papers and partially
integrated into some KL1 programs, but mostly existing in the brains
of the researchers.
Results have not been well ordered yet and thus are very abstract.
Experience on how to represent legal knowledge gained in the
development of the legal reasoning system is another abstract result.
To disseminate these abstract research results, we decided to set
up several small collaborative projects with Japanese universities. In
these projects, professors are expected to use some of our results as
seeds in starting new research projects. Of course, they are asked to
use practical results as educational tools or as new infrastructures.
Currently, we have the following 15 projects.
- Parallel language processors and environments;
- Optimization of a language processor (Prof. H. Tanaka, Univ. of Tokyo)
- Optimal implementation of KL1 (Prof. H. Nakashima, Kyoto Univ.)
- Program analysis and optimization (Prof. K. Ueda, Waseda Univ.)
- Visual interface of parallel systems (Prof. J. Tanaka, Tsukuba Univ.)
- Parallel theorem proving;
- High-speed inference (Prof. K. Fuchi, Univ. of Tokyo)
- Lannguage processor based on theorem proving (Prof. M. Amamiya, Kyushu Univ.)
- Natural language processing;
- NL processing on the PIM (Prof. H. Tanaka, Tokyo Institutes of Technology)
- Parallel NL understanding (Prof. R. Taniguchi, Kyushu Univ.)
- NL tools (Prof. Y. Matsumoto, AIST-Nara)
- Others;
- Distributed AI systems (Prof. K. Furukawa, Keio Univ.)
- Constraint processing (Prof. F. Mizoguchi, Science Univ. of Tokyo)
- Parallel VLSI CAD systems (Prof. K. Taki, Kobe Univ.)
- Evaluation of parallel architectures for LP (Prof. H. Tanaka, Univ. of Tokyo)
- Transaction management (Prof. H. Yokota, JAIST-Hokuriku)
- Methodology for parallel scientific computation (Prof. S. Kunifuji, JAIST-Hokuriku)
Furthermore, we have two collaborative projects with the
Electrotechnical Laboratory (ETL) and the Mechanical Engineering
Laboratory (MEL). These are national laboratories which belong to
MITI. These projects are continuations of the FGCS Project.
3.3 Collaboration on international research
and achievement sharing
The technical goal of the FGCS Project was very advanced and
needed the collaboration of the world's leading researchers. Thus,
international collaboration began in the initial stage of the FGCS
Project. Early collaboration was based on invitation or mutual visits
of individual researchers.
From the intermediate stage, we concluded formal agreements for
exchanging researchers and holding workshops with the following
governmental organizations;
- National Science Foundation (NSF) in the U.S.A. (Mr. Y.T. Chien)
- Institute National de Recherche en Informatique et en Automatique (INRIA) in France (Dr. L. Kott and Dr. G. Kahn)
- Swedish Institute for Computer Science (SICS) (Dr. S. Haridi and Dr. M. Nilsson)
- Department of Trade and Industry (DTI) in the UK (Dr. K. Shotton and Dr. P. Rothwell)
In the final stage, we started more substantial research
collaborations to find appropriate applications for the KL1, PIMOS and
PIMs. We exchanged ideas and research tools with the following U.S.
laboratories;
- Argonne National Laboratory (ANL) (Dr. E. Lusk, Dr. R. Overbeek and Dr. R. Stevens)
- National Institute of Health (NIH) (Mr. R.J. Feldmann and Dr. G.S. Michaels)
- Lowrence Berkeley Laboratory (LBL) (Dr. C. Cantor and Dr. C. Smith)
With researchers at ANL, we conducted research on biological
analysis and theorem proving. They gave us several interesting
problems on theorem proving. They greatly stimulated our research on
parallel theorem proving and biological analysis. We held US-Japan
workshops on logic programming and theorem proving twice at ANL
sponsored by NSF and ICOT.
With researchers at NIH and LBL, we conducted research on protein
structure analysis and sequence alignment. We learned much about
molecular biology through these collaborations.
The collaborative research on theorem proving was extended to
include Australian researchers.
- The Australian National University (ANU) (Prof. M. McRobbie and Dr. J. Slaney)
ANU worked with us to try to solve some open problems in the
quasi-group theory using the MGTP theorem prover on a PIM. This trial
was very successful. The achievements of this collaboration
stimulated the world's researchers in this area very much. We held
several international workshops with these researchers to share our
research results and experiences.
These collaborations were really helpful to our growing research
groups into genetic information processing and parallel theorem
proving. Most of these collaborations are still continuing on an
individual basis.
In the Follow-on Project, we set up new tighter collaborative
research projects with the following two universities;
- The University of Bristol (Prof. D. Warren and Prof. S. Gregory)
- The University of Oregon (Prof. J. Conery and Prof. E. Tick)
We hoped to further develop the KLIC system and application
systems as well as the disseminating the KLIC system and other
programs as ICOT Free Software.
With researchers at the University of Bristol, we are working to
develop a parallel debugger and a constraint solver which can be used
as extended functions in the KLIC system.
With researchers at the University of Oregon, we are conducting
research on optimizing the compiler for KLIC and biological sequence
analysis using constraints. Based on this collaboration, we held the
fifth US-Japan workshop at this University attended by international
researchers.
The results of all these collaborations will be placed in the
public domain and distributed as extensions of ICOT Free Software.
With the completion of the Follow-on Project, we have to terminate
all of these collaborations. We hope that the work will continue on
individual bases.
3.4 Distribution of ICOT Free Software (IFS)
and other technical information
We started the distribution of ICOT Free Software (IFS) using
anonymous FTP on the Internet in August 1992. Since then, 12,000
files have been transferred to 2,200 sites in 45 countries. Among
these file transfers, 40 % were to Japan, 30 % to the U.S.A. and the
remaining 30 % to other countries.
In the final year of the FGCS Project, we placed 77 programs in
the public domain as IFS and started up ICOT's FTP server in August
1992. These programs are relatively large. For example, the PIMOS
operating system is counted as one program. These programs include
many large programs written in KL1 which would not run without a PIM.
Thus, we wondered if there would be any interest from researchers
outside of ICOT. Fortunately, the IFS server was frequently accessed
from many countries from the beginning.
In the Follow-on Project, we made great efforts to complete the
sequential version of KLIC quickly so that IFS users could have an
environment to be able to run KL1 programs on their machines. We
released 7 programs as IFS including the sequential version of KLIC in
November 1993. As we expected, the KLIC was frequently transferred by
many people.
In March 1995, as the final release of IFS, we will add 16
programs to IFS including the parallel version of KLIC and several
interesting parallel application programs. They are actually
operational on many workstations and parallel machines.
In October 1994, we started up ICOT's World Wide Web server. We are
providing general information on ICOT's activities, ICOT publications
such as technical reports and ICOT journals, and outlines of major
research results. ICOT Free Software is also accessible from this
server.
Since its start up, the ``ICOT Home Page'' file has been
transferred to more than 800 sites. This is very encouraging. We plan
to put many more recent achievements on the server.
By using the Internet, we were able to distribute ICOT Free
Software and other technical information quite effectively. Although
face to face communication is indispensable for successful
collaborations, we naturally rely more on computer networks for
sharing advanced technical information, research tools, and products.
We plan to organize a team of researchers and engineers to
maintain ICOT Free Software and the other information for a few year
after the ICOT Research Center closes in March 1995.
The general goal of the Follow-on Project has two aspects; One
is to disseminate the achievements of the FGCS Project. The other is
to make a "soft-landing" for the FGCS Project which has been
Japan's largest national project in the area of computer technology.
We are sure that we can attain our goal for dissemination.
First of all, the development of KLIC is proceeding smoothly. The
KLIC sequential version is being smoothly ported to Unix-based
workstations by many KLIC users outside of ICOT. The number of KLIC
users is increasing rapidly both in Japan and other countries. They
use KLIC mainly for research and educational purposes.
The KLIC parallel version is now being used at ICOT and is being
debugged. The KLIC parallel version is also being ported to several
parallel machines at several different sites.
Porting is rather easy. However, to get the best performance,
optimization using dedicated hardware functions for parallel machines
is necessary.
We expect that this problem will be solved because middle-level
software which supports parallel processing such as PVM is now rapidly
being optimized on many parallel machines.
Improvement and porting of the application systems are still
underway. One of the problems we have in this porting is that large
scale MIMD parallel machines have not yet become popular so that many
researchers do not have easy access. Particularly the number of
element procesasors and memory capacity is not large enough for us to
port our application systems.
The expectations of users of the parallel machines seems to be
very large. However, parallel machine vendors feel that the market is
still small. We understand that this gap is caused by the lack of an
efficient parallel programming language and environment. We are
confident that KLIC is a promising solution to fill this gap.
We will also accomplish the goal of a soft-landing for the FGCS
Project. We are glad that ICOT researchers are getting better
opportunities and better positions in academia and industray.
Many of the ICOT graduates who returned to their companies are
engaged in interesting jobs such as development of parallel machines
and distributed software. ICOT researchers who were at ICOT for
several years and led the FGCS Project are welcomed by many
universities.
We are happy that ICOT's PIMs have been welcomed by several
universities. They will be moved to these universities and used to
foster researchers for the next decades.
To maintain and further disseminate ICOT Free Software, we plan to
organize a virtual research laboratory on the Internet comprising ICOT
graduates and collaborative research partners. This will be a
loosely-coupled research group who share a common interest in ICOT
Free Software. We hope that this group will act as a think-tank to
foster future research projects.
- Chikayama 1994
- T. Chikayama,
``Parallel Basic Software'', Proceedings of the International
Symposium on Fifth Generation Computer Systems 1994, Dec. 1994.
- Nitta et al. 1994
- K. Nitta, K. Yokota,
A. Aiba and M. Ishikawa, ``Knowledge Processing Software'',
Proceedings of the International Symposium on Fifth Generation
Computer Systems 1994, Dec. 1994.
- DPBCT 1994
- Committee for Development
and Promotion of Basic Computer Technology, ``Fifth Generation
Computer Systems Project Final Evaluation Report'', ICOT Journal, No.
40, pp-2-24, 1994.
- Uchida et al. 1993
- S.
Uchida, T. Chikayama, and K. Nitta, ``Knowledge Information Processing
by Highly Parallel Proccessing'', ICOT Technical Report TR-0854, Sep.
1993.
- Uchida et al. 1993
- S. Uchida, R.
Hasegawa, K, Yokota, T. Chikayama, K. Nitta and A. Aiba, ``Outline of
the FGCS Follow-on Project'', New Generation Computing, Vol. 11, No.
2, OHMSHA and Springer-Verlag, 1993.
- Uchida (Ed.) 1992
- S. Uchida (Ed.),
``Proceedings of the FGCS PROJECT EVALUATION WORKSHOP'', ICOT
Technical Memo TM-1216, Sep. 1992.
All of this proceedings (Compressed PostScript file) :
General Report of the FGCS Folow-on Project (68KB)