[comp.research.japan] Trip Report, IBM Tokyo Research Laboratory

rick@cs.arizona.edu (Rick Schlichting) (02/05/91)

--------------------------------------------------------------------------
    Trip Report -- IBM Tokyo Research Laboratory  (July 5, 1990)

    Richard D. Schlichting   (rick@cs.arizona.edu)
    Associate Professor
    Department of Computer Science
    The University of Arizona
    Tucson, AZ  86721, USA 

    [This report represents the personal opinion of the author, who was on
     sabbatical in Japan from Dec. 1989 through July 1990.  The sabbatical
     was supported in part by grant INT-8910818 from the NSF U.S.-Japan
     Cooperative Science Program.]
---------------------------------------------------------------------------

    The day following my visit to Hitachi, I traveled to IBM TRL for
my last visit during this stay in Japan.  The arrangements were made
with Dr. Norihisa Suzuki, the director of the lab.  David Notkin
from the University of Washington was also arranging a visit at the
same time, so we ended up going together and giving back-to-back talks.
Since David had visited the previous year and had research interests
more in line with the work going on there, I had the luxury of sitting 
back and listening more than on some of my other visits.
    David and I arrived at the lab at about 12:45, where we were met 
by Dr. Toru Takeshita.  Dr. Takeshita is Manager of the Computer Science 
Institute of the lab and the author of a recently completed book 
(in Japanese) on CASE (Computer Assisted Software Engineering.)  We 
chatted in the lobby about various things for about half an hour (including 
how important Japan is to IBM -- apparently it accounted for 10% of total
worldwide profits in the most recent fiscal year) , at which time we
headed to a nearly building where the actual talks were to be held.
My talk was first, and lasted about an hour; Dr. Suzuki entered a
few minutes before it finished.  David talked second, and put us
back on schedule by ending in about 45 minutes.  
    The next thing on our agenda was a series of discussions and
presentations.  The first thing was a presentation by Tamiya Onodera
on the COB programming language project.  (The head of the project
was out of town.)  COB is an object-oriented version of C that solves 
some of the problems associated with C and C++.  As was remarked several 
times, (by both them and us), they are probably fighting a somewhat 
losing battle despite a better product due to the widespread use of C++; 
as they said, they really started a couple of years too late to make a 
big impact.  Nevertheless, they have some interesting results and were 
able to give a nice demonstration of a development environment system for 
COB including a debugger that shows a dynamic graphical representation 
of the program during execution.  As David remarked, the debugger is 
a nice example of a different kind of program animation.
   We then returned to the conference room, where Takeshita gave us
an overview of the TRL and its place within the IBM corporate structure.
The number of researchers in the lab is about 260, which compares to
2400 at TJ Watson and 750 at Almaden. The four specific areas covered
by TRL are Computer Science, Computer Applications, Component
Technologies, and Manufacturing Technologies.  He said that TRL is
considered especially strong in the applications field.  In that
area, they currently have efforts in graphics, application development
environments, man-machine interfaces, natural language processing, image
processing, knowledge-based systems, and prototyping.  On natural language
processing, he indicated that their English to Japanese translator is
unique since it first rearranges the English into a more Japanese word
order, and then translates to Japanese.  The result, he said, is 70% good
and is currently used for translating such things as technical manuals.  
In the CS area, there are projects in systems (architecture, OS, programming 
languages and environments, software tools, the TOP-1 multiprocessor, 
and multiprocessor AIX) and theory (theoretical CS and OR.) Takeshita then 
gave us a talk on a description and history of Japanese software development 
efforts, with an emphasis on CASE.  
    The next talk was a description of the TOP-1 multiprocessor project.
This presentation was made by Dr. Nagatsugu Yamanouchi, who is manager 
of the workstation software group.  The TOP-1 project, which is well-known 
in Japan, started in 1987, with the hardware prototypes being finished 
in 1988 and the OS in 1989.  The machine is a shared memory multiprocessor 
consisting of 10 386-type multiprocessors.  Standard memory configuration 
is 32 Mb, but it can take up to 128 Mb.  Claimed as one of its unique 
features is its snoopy cache, which consists of 128 kB of storage per 
processor; the cache uses a dynamic variant of standard snoop protocols,
and is switchable between either of two modes on a per processor basis. 
Another unique feature is that the hardware is asymmetric since the disk 
subsystem is only accessible to only one processor.  As a result, the OS 
has also been constructed asymmetrically, with the one processor that can 
communicate with the disk dedicated to executing the kernel and the others 
running user tasks.  Also, the system has been designed with a built-in
hardware monitor to facilitate collection of various statistics, something
that has not been done before with this kind of architecture. The machine
was originally targeted for numerical applications, although it doesn't 
really enough horsepower to be useful currently and the asymmetric hardware 
and software architecture would seem likely cause bottlenecks if the 
number of processors were expanded significantly.  Two languages are 
currently being used: a Multiprocessor Lisp developed at TRL and a 
MP Fortran that is apparently based on the IBM parallel Fortran developed 
at Yorktown Heights.  Ten prototypes of the system were built, with four 
currently on loan to Japanese universities (specifically, Keio, Waseda, 
Tokyo, and Kyoto.)
    At the end of all this, we were taken to Suzuki's office, where
we sat and chatted for awhile.  He talked quite easily and freely
about the projects in his lab, and seemed very interested in the
comments that we offered about the COB and TOP-1 projects.  He
clearly supports the type of basic research represented by these
projects, but with the recognition that part of his job is to
ensure that the company's money is used productively.

--------------------------------------------
Dr. Norihisa Suzuki
Director, IBM Tokyo Research Laboratory
IBM Japan, Ltd.
5-19 Sanbancho, Chiyoda-ku
Tokyo 102, Japan

Dr. Nagatsugu Yamanouchi
Manager of Workstation Software
IBM Tokyo Research Laboratory
IBM Japan, Ltd.
5-19 Sanbancho, Chiyoda-ku
Tokyo 102, Japan

Dr. Toru Takeshita
Manager of Computer Science Institute
IBM Tokyo Research Laboratory
IBM Japan, Ltd.
5-19 Sanbancho, Chiyoda-ku
Tokyo 102, Japan