Kahaner Report -- Indian parallel processing activities [abstract only reset on request. BTW the Ind

[David Farber <>]

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: Indian parallel processing activities
04/14/94 (MM/DD/YY)
This file is named "india-pp.94"


ABSTRACT.  A summary of several indigenous Indian activities in parallel
processing. There are five or six independent projects to construct
parallel processing systems. Efforts are being made to commercialize
some of them. The projects combine a learning function for Indian
scientists, with a response to difficulties that India has (financial
and political) in acquiring advanced computing from outside the country.


During my first trip to India I visited the Center for Development of
Advanced Computing (CDAC), as well as the National Aerospace Lab (NAL)
and reported on their work in parallel computing, see my report
"india.93a" 28 Jan 1993 for details. During a recent visit I revisited
CDAC as well as several other Indian parallel processing projects. These
are described below. A second report, in progress, describes other
computer related R&D in India.


The focus of my visit was participation in a conference "Seminar on
Supercomputing for Scientific Visualization (SSV'94)", held at the
Bhabha Atomic Research Center (BARC), and jointly organized by the
Computer Division at BARC and the Indian Physics Association.  An opening
day of tutorials, three days of invited and contributed talks, as well
as a poster session and small exhibit were attended by several hundred
scientists. In what follows I summarize six parallel processing
projects, and also discuss relevant papers. A complete list of papers
and authors is given at the end of this report.


India has need for advanced computing, much as other countries do. Areas
specially identified are as follows. While there are no surprises, the
selection of topics is interesting.


      Computational fluid dynamics
      Design of large structures
      Computational physics and chemistry
      Climate modelling
      Vehicle simulation
      Image processing
      Signal processing
      Oil reservoir modelling
      Seismic data processing.


Major Indian parallel computing projects.


PARAM (from Center for Development of Advanced Computing--CDAC, Pune)
ANUPAM (from Bhabha Atomic Research Center--BARC, Bombay)
MTPPS (from Bhabha Atomic Research Center--BARC, Bombay)
PACE (from Advanced Numerical Research Group--ANURAG, Hyderabad)
CHIPPS (from Center for Development of Telematics--CDOT, Bangalore)
FLOSOLVER (from National Aerospace Laboratory, Bangalore)


All India's parallel processing efforts share some common historical
characteristics. (1) Government organizations have severely restricted
foreign exchange budgets but a large, highly educated professional
work force.  (2) Because of COCOM rules, Indians are denied access to
advanced computing technology from the West or Japan, except in very
controlled situations. (There is one Cray X-MP used for weather
modelling at the Center for Medium Range Weather Forecasting, NCMRWF,
which is under very tight security. An effort to purchase a newer Cray
at the Indian Institute of Science in Bangalore was ultimately abandoned
partly because of the tedious negotiations and security requirements,
and partially because of the emergence of indigenous Indian parallel
products that were seen as partial substitutes.) These characteristics
have led to a variety of efforts to develop indigenous parallel
computers utilizing commodity processors. Almost all the efforts began
with transputers as the basic cpu, because of their accessibility and
ease of implementation.  Lately these systems have been moving up-scale,
to faster transputers, and also to i486 and i860 processors. There is
even some effort to utilize SPARC or RISC processors, and all the
designers have recognized that a processor independent design is
required in order to keep up with changing technology. System and
application software is being developed locally, as is documentation and
support service. Most of the projects are confidence building exercises,
training for large numbers of Indians in parallel computing, or
developing from the requirements for running specialized local
applications in the nuclear, weather, aeronautics, and other fields.
While all the efforts would like to spin off into commercial products,
only one or two are well on their way, CDAC's PARAM being the furthest
along the business path, with BARC's ANUPAM also having a good
technology and software base.  Given India's low labor cost, it is quite
possible that one or more of them can be developed into a product that
will be successful selling into certain markets. Or, as one Indian
scientist commented, that Indian  parallel processors could be the
cheapest, although not necessarily the best.


Several groups are experimenting with parallelizing the source for one
of the weather modelling programs that is now running on their Cray.  At
this time, none of the existing platforms has generated sustained
performance on real applications much beyond the X-MP range, although
some code kernels are doing better. But there are serious efforts to
improve performance by increasing the number of processors, using faster
processors, and tuning the software.  At the very least the development
of these Indian systems has generated a body of local scientists at the
government labs who are quite well versed in parallel computing, at
least using commodity processors, and who have had opportunities to
experiment with a variety of applications.  As one indication of this
activity, the Supercomputing for Scientific Visualization (SSV'94) was
held at BARC in Bombay Feb 14-17, 1994, for which the theme was "think
parallel". Seventy seven  papers by Indians, more than 90% from authors
entirely within the country, were presented.


Although there are some scientists who would like to give up the efforts
to develop Indian parallel computers, most accept that they will have to
trade their own labor and convenience as a replacement for costly
commercial systems. The keyword from many scientists I met was the use
of "existing resources", a euphemism for "doing it ourselves."


While traveling in India, I heard little enthusiasm voiced for buying an
expensive Western supercomputer. Even the weather people who are running
the X-MP (which is now getting old and running out of steam), while
admitting that there are significant differences between research and
operational requirements for computing, recognized that parallel was the
future and that the most cost effective approach was to trade
(inexpensive) Indian labor for finished Western products. However, if an
organization is able to get a high performance computer installed, there
is a great deal of enthusiasm about using it. For example,


     Dr.K.S.Yajnik, Head
     Center for Mathematical Modelling and Computer Simulation (C-MMACS)
     National Aeronautical Laboratory
     Bangalore 560 017, India
      Tel: +91 812 574649, 579611 x411; Fax: +91 812 560392
      Email: YAJNIK () CMMACS ERNET IN
has just installed a new Convex 3810, and was excited about being able
to run many well known software packages on a stable computing platform;
C-MMACS also has several DEC Alphas and SGI Indigos.  There is also a 16
node older parallel system developed locally, but it is not used much
because the CPUs are too slow.


This points up a conundrum facing Indian scientists, and mentioned to me
several times. Without a strong application software base on Indian
parallel computers, individual scientists need to spend a great deal of
time porting programs. What many have discovered is that at the end of
six months or a year of work, total problem solution time on their
parallel machine may not be much better than that on the newest class of
workstations. The Indian machine designers realize this, and have been
trying to develop more processor independent systems, as well as work
diligently to port applications. However, without a tight relationship
with cpu developers (in the West) they find it difficult to get early
information and delivery of the newest processors; once behind that wave
it is difficult to catch up. Another possible approach is to leverage
cheap Indian labor and software expertise by collaborating with Western
software vendors to get applications running quickly on parallel systems.
This is being done to some extent but projects need to be carefully
structured to allow work to be done by teams remote from the mother
company. A problem mentioned to me in this regard is the concern that
small Western software companies have about maintaining control of their
intellectual property (i.e., source programs). Ultimately, mutual trust
and a positive track record are essential. A third possibility is to
develop all software in-house. To a large extent this is what is being
done now, and Indian scientists have a great deal of traditional
software experience but this approach seems to have plenty of its own
problems and implications.


Incidentally, I also did not notice much interest, or knowledge, about
Japanese developments in high performance computing, and generally one
does not see as much evidence of Japanese activity in India as in other
Asian countries. Similarly, back in Japan, there seems to be a sense
that the Indian market is for the future. (This may not be true in other
technology areas. For example, there is a major automotive collaboration
between Tata and Suzuki.)