ISR's High Performance Computing & Networking Workshop

[David Farber <farber () central cis upenn edu>]

From: 
 Dr. David K. Kahaner
 US Office of Naval Research Asia
 (From outside US):  23-17, 7-chome, Roppongi, Minato-ku, Tokyo 106 Japan
 (From within  US):  Unit 45002, APO AP 96337-0007
  Tel: +81 3 3401-8924, Fax: +81 3 3403-9670
  Email: kahaner () cs titech ac jp
Re: ISR 1993 Workshop
8 Sept 1993
This file is named "isr.93"


ABSTRACT. ISR's High Performance Computing & Networking Workshop,
Tsukuba Japan 8/93.


ISR, originally called the Institute of Supercomputing Research, when it
was part of the Recruit company, has had a research lab in Tokyo for
several years. Several Western scientists have been involved in
collaborative research there, for example, Dr. H.Wasserman from Los
Alamos, and others. ISR also published the Vector Register, a useful
newsletter of activities in Japanese high performance computing. In
addition, ISR has had well received annual workshops on high performance
computing for the past five years (see for example, "workshop.isr", 27
Sept 1990).


However, Recruit has decided that high performance computing does not
fit with their other business interests and during the past year the
facility has been transformed into a multimedia development center.
ISR's Director, Dr. Raul Mendez, has recreated ISR as a private
organization, International Systems Research, without ties to Recruit.
In this activity he is joined on the US side, by Mr. Craig Lund, who
runs Local Knowledge from Durham NH. Their initial activities will be
associated with system integration, with particular interest in the
short term, on the 11 supercomputer systems that are to be purchased and
installed by April 1994 at Japanese government laboratories as part of
the economic stimulus package.


For information about ISR, contact
             ISR
             Villa Pinedo 103
             Akasaka, Minato-Ku
             Tokyo 107 Japan
              Tel: +81 3 3423-9509; Fax: +81 3 3423-0979


Last month, ISR ran its annual workshop, in Tsukuba. But in keeping with
the new organization's charter, the emphasis was on bringing users and
potential users together with vendors. As distinguished from past ISR
workshops, this was a not a research forum. Instead, six speakers gave
tutorial, or overview talks. In addition, there was a small exhibition,
with five vendors represented, NEC, Cannon-KSR, IBM, HP, and
Matsushita-Parsytec.  (From the Japanese side, it was very interesting
to see NEC's material on the Cenju-3, a multi-stage net connected
parallel machine with up to 256 50MFLOP VR4400SC processors.) The
workshop was also co-sponsored by the Ministry of Education's Science in
Education Research Foundation. Supporting organizations included the US
Embassy and the journal Parallelogram.


Approximately 100 actual participants (excluding vendors) were present,
and ISR did a good job of getting representation from all the 11
government sites planning supercomputer purchases. In past years, a
large conference and exhibition, Japan Supercomputing 199x, was held, but
the 1993 conference was cancelled and I am told that the planned 1994
conference is also likely to be similarly cancelled due to high costs.
Thus the ISR workshop was one of the few opportunities for potential
users to come together and hear about systems. Although there were a few
participants who were very well versed in supercomputing, many were
clearly here to learn, and in that sense they were well served and the
workshop should be deemed a success.


Below I have included summary material on the Japanese presentations.
(Two presentations, in English, were overviews of activities in the US
on clustered workstations and High Performance Fortran.) Of these, the
most interesting for me, and perhaps for Western readers, was the
overview of NTT's networking and service strategy, called by the company
VI&P (Visual, Intelligent and Personal Communications). This is related
to applications of ISDN.  As I have written on this topic several times
before, here I will only summarize the latest activities, what NTT calls
Phase II.  (For associated material, both narrow and broad band ISDN,
see the reports, "j-comm.93" 5 Sept 1993, "isdn.93", 25 Feb 1993, and
"isdn.92" 6 May 1992.) At the ISR workshop, the NTT presentation was
made by


             Dr Mitsuru Miyauchi
             Senior Research Engineer, Supervisor
             Network Integration Laboratory
             NTT Telecommunication Networks Lab
             1-2356, Taki, Yokosuka-shi 
             Kanagawa-ken 238-03 Japan
              Tel: +81 468-59-3180; Fax: +81 468-59-2546
              Email: MIYAUCHI () NTTMHS NTT JP




Since April 1991, NTT has been conducting comprehensive experiments
related to VI&P (Visual, Intelligent and Personal communication
services), NTT's service vision for the 21st century at its company's
laboratories.   B-ISDN and FTTH (Fiber to The Home) are important and
basic items for the VI&P concepts, and are being developed in the NTT
laboratories. Existing ISDN experiments at NTT have focused on
optimizing the use of current N-ISDN (called by NTT, INS-Net). Beginning
this fiscal year, the company is concentrating on experiments for new
services and applications for the time when higher bandwidth becomes
available.  This summer, NTT  released information about their work in
two general areas, (1) multi-media and (2) high-speed computer
communications.




(1) Multi-media.
NTT's Yokosuka R&D center (about one hour south of Tokyo) is connected
by an optical subscriber transmission system to a "house" about 4km
away.  The system can multiplex-transmit data at speeds of several Mbps
up to several hundred Mbps on a single optical fiber.  The environment
in the house is supposed to simulate a real home or office situation, so
in addition to sophisticated issues, NTT is also studying optical fiber
cable wiring questions such as the installation of optical cables,
mainly applied indoor. NTT plans to install FTTH (Fiber To The Home) by
year 2015 in Japan. Experiments in the house are related to simultaneous
use of multi-services and applications of super-high-definition, (SHD)
(see below) images.  Specific applications include electronic newspaper
transmission, and transmission of photomicrographs of cellular tissues
or X-ray data for medical diagnosis.


Experiments related to simultaneous multiple services involve the use of
a video telephone and 60 channels of TV (including some HDTV) on a
single optical fiber using the optical subscriber transmission system
mentioned above. SHD applications (see next paragraph) in the office are
envisioned to include such things as fabric databases, high definition
images from libraries and museums, CAD, etc.


NTT's SHD (Super High Definition) is a system to transmit and display
images with four time HDTV data capacity (claimed to be about the
resolution of 35mm film). Using 150Mbps (B-ISDN) and this system, NTT
can transmit a still image in less than one second. Using 600Mbps lines,
a high-quality moving picture can be transmitted in real time.


(2) High-speed computer communications.
The Yokosuka R&D center is also connected to NTT's Musashino R&D center
via ATM links connected to a 100km 2.4Gbps optical transmission line.
(Musashino is west of Tokyo.)  The centers will be also connected by
analog NTSC and digital 156Mbps satellite links in 1993 and 1995
respectively. The idea is to study problems associated with various
applications including latency and error effects. Within each center
FDDI LANs are installed. There is also a supercomputer at Musashino
which generates computer graphics image data sent to an HDTV quality
display at Yokosuka. There are also multimedia system experiments such
as PC-based multipoint desktop video conferencing using ATM-LAN
developed at the NTT labs.




Remark: Readers of my reports will doubtless get the impression that
there is some disarray in Japan's activities with regard to ISDN. For
example, the report "j-comm.93" 5 Sept 1993, suggests that financial
pressures may retard the completion dates of the FTTH project. The
Japanese economy is not nearly as strong as in the past, competition
is intense within the communications industry, and there are many
governmental "turf wars" -- nevertheless, forward progress is evident.


------------------------------------------------------------------------


  Simulation Supercomputer Center at National Institute for Fusion Science
             Kunihiko Watanabe, NIFS


This talk described the newly created supercomputer center at the
National Institute for Fusion Science in Nagoya. Their "Simulation
Analysis System" is centered on an NEC SX-3 24R supercomputer that a
small number of users share. (The decision to obtain this machine over a
Cray caused front page headlines here several months ago.) The basic
concept is to try to advance the understanding of fusion physics by
increasing the computational resources dedicated to numerical
simulation. The NIFS system is unique in several ways (even by Japanese
standards).  For instance, it is designed for handling 4GByte, 2-hour
jobs in single user mode. Notable  features of the system are its use of
mass storage (1.5TByte, 10MByte/sec throughput) as disk space and its
use of a 160Gbit RAID disk system as a buffer from which data is
transferred to mass storage after job execution.




------------------------------------------------------------------------
Amateur's Experiences on a Highly Parallel Computer, A Large Scale
  Numerical Simulation on AP-1000 (QCD-TARO Project).
      Atsushi Nakamura, Faculty of Education, Yamagata University


This talk described the author's experience with parallel processing on
the Fujitsu AP-1000, a distributed memory system based on the Sparc RISC
chip. The AP-1000, a 2-D mesh (or more precisely a torus), is the
parallel processor with the largest number of users in Japan and this
talk was a kind of user's log of experiences porting QCD applications to
the AP. The mapping from the physical domain to the topology of the AP
as well as load balancing in QCD applications were discussed in detail.
(TARO stands for Thousand cell Array processor for Omnipurposes.)


-----------------------------------------------------------------------


              The Emerging Workstation Cluster
                Craig Lund, ISR America, Durham NH


-------------------------------------------------------------------------


              HPF: Why, Where, and How
                Larry Meadows, The Portland Group, Wilsonville OR


------------------------------------------------------------------------
              
              Benchmarking, Past, Present, and Future
                S. Sekiguchi, ETL, Tsukuba Japan


This talk surveyed the performance of supercomputer systems. A review
was first given of the various attempts at establishing a methodology
for measuring performance. The author emphasized that benchmarks should
be used for performance estimation or prediction when users purchase new
machines or when designers develop new machines.  Thus, one of the
important features of a benchmark is that it must cover the perspectives
of both applications and architectures. Of course, "performance" is
supposed to be a function of computing something, and hence, once an
application has been run, the output should contain the data necessary
for evaluation.


The PERFECT, NAS, SLALOM, SPLASH, Euroben, and several other benchmarks
were discussed. These varied from the hierarchical approach, proceeding
from simple vector operations, to kernels, advancing to subroutines, and
reaching up to full application programs. There are mutual interactions
among a benchmark's hierarchical levels.


Sekiguchi also stated that MITI is planing to start an academic project
for research on computers' performance modeling and its theory.