Newsgroups: comp.sys.super
Path: sparky!uunet!think.com!ames!data.nas.nasa.gov!orville!eugene
From: eug...@orville.nas.nasa.gov (Eugene N. Miya)
Subject: So the news says....
Sender: ne...@nas.nasa.gov
Organization: NAS Program, NASA Ames Research Center, Moffett Field, CA
Date: Thu, 13 Feb 92 18:14:28 GMT
Message-ID: <1992Feb13.181428.4438@nas.nasa.gov>
Lines: 15
That IBM is formally getting into the supercomputer market.
Time to suspend IBM jokes and see what they can do.
Details?
Which Division? What Lab? Who are the architects? What approaches are they
they taking to achieve balance? What specific application area is
driving their development (surely not the DOD/DOE nuclear weapons community).
Software (will they run MVS & parallel COBOL [ooops! sorry.])?
What about mass storage (rates and volumes)? Time tables? Do I have
to sign more non-disclosure to learn this (sure will)? Will they change their
policy on benchmark disclosure?
--eugene miya, NASA Ames Research Center, eug...@orville.nas.nasa.gov
Resident Cynic, Rock of Ages Home for Retired Hackers
{uunet,mailrus,other gateways}!ames!eugene
Xref: sparky comp.parallel:896 comp.sys.super:431
Newsgroups: comp.parallel,comp.sys.super
Path: sparky!uunet!europa.asd.contel.com!gatech!hubcap!fpst
From: eug...@nas.nasa.gov (Eugene N. Miya)
Subject: Re: So the news says....
Message-ID: <1992Feb14.162045.550@hubcap.clemson.edu>
Sender: fp...@hubcap.clemson.edu (Steve Stevenson)
Organization: Clemson University
Date: Fri, 14 Feb 92 08:13:50 -0800
Approved: para...@hubcap.clemson.edu
Lines: 153
From: MCIN...@CERNVM.CERN.CH
Subject: Re: So the news says....
To: "Eugene N. Miya" <eug...@wilbur.nas.nasa.gov>
Hello again; for what it is worth here is what I received..........
I must say we do find the RISC 6000 very good performance for money
and it will be interesting to watch CRI+DEC alpha v IBM RISC.
Regards. Eric.
----------------------------Original message----------------------------
I am pleased to include a copy of today's announcement of IBM's
new supercomputing strategy. I hope it will be of interest to
you and I look forward to answering any questions you may have
about it.
Best regards, Tito Sorlini
*******************************************************************
IBM ANNOUNCES NEW SUPERCOMPUTING STRATEGY
PARIS . . . February 13, 1992 . . . In a wide-ranging
statement of direction, IBM today announced a new supercomputing
strategy, presented before a Paris audience at Super Computing
Europe '92. The announcement focuses on the creation of a new family
of highly parallel supercomputers which will be brought rapidly
to market by a recently formed supercomputing laboratory.
The Highly Parallel Supercomputing Systems Laboratory
(HPSSL), based in Kingston, N.Y., will design, develop and
deliver a series of scalable parallel supercomputing systems, using IBM's
industry-leading RISC System/6000* technology. This move complements
work already performed by IBM in Europe to develop solutions for
parallel computing based on clusters of IBM RISC System/6000s,
which are available now.
"We are proud of the success record that IBM's pioneer RISC
technology has established," said Jack D. Kuehler, President, IBM.
"Success breeds success and we look forward to expanding this technology
into new systems capable of meeting our customers' ever-increasing
appetite for more and more computational power."
The new development effort combines IBM's resources
from several business units, including the Enterprise Systems
line of business, the Advanced Workstations Division, the IBM
Research Division and the Federal Sector Division. The group will use
multiple UNIX**-based RISC processors to create a scalable,
highly parallel system capable of performance in the range of
hundreds of gigaflops.
The architecture and design are intended to achieve teraflop
performance.
The resulting products are intended for scientists, engineers,
researchers and analysts who require greater processing power to
solve increasingly complex problems such as those found in:
computational fluid dynamics, financial modelling, weather
forecasting, computational chemistry, pharmaceutical design,
seismic data analysis, reservoir modelling, structural analysis
and engineering design in the automotive and aerospace industries.
"Our key customers have told us they need scalable
systems that enable them to build up performance on an
incremental basis - as their individual needs require," said
Irving Wladawsky-Berger, assistant general manager for
supercomputing, Enterprise Systems line of business in IBM.
"We will package the hardware and software technology
to give our customers as much performance as they demand
and we plan to get these machines to market as quickly as
possible."
The highly parallel systems being developed by the new
supercomputing laboratory are part of IBM's multi-level
supercomputing strategy which includes:
* Continued enhancement of the IBM vector facility, an
optional feature for numerically-intensive applications
which is available on ES/9000* and ES/3090* systems. More
than 500 of these systems are installed worldwide.
* The development of a standalone highly parallel system,
using large numbers of RISC-based processors in a single
scalable system capable of hundreds of Gigaflops, which
can optionally be integrated with ES/9000 processors.
* IBM RISC System/6000 clusters consisting of 3 to 32
economical RISC System/6000s, which already provide an
entry-level parallel server, batch server and data server
solution.
* Development alliances with other companies that are
intended to complement the above offerings.
Initial delivery of the first low-end system being developed
by the IBM Highly Parallel Supercomputing Systems Laboratory
is expected to be announced later in 1992. Follow-on
systems, with additional numbers of processors, will be offered on
a regular basis throughout the 1990's.
The new scalable parallel systems will build on the experience
gained with the parallel RISC System/6000 solutions
developed by IBM at the Rome European Centre for Scientific and
Engineering Computing (ECSEC) and the Stavanger European Petroleum
Application Centre (EPAC). These solutions exploit the use of multiple
RISC System/6000s for parallel and batch processing, as well as closely
coupled input/output processing with ES/9000 systems.
They are available now throughout Europe.
Dr. Herbert Budd, IBM's European Director of Scientific and
Technical Solutions, presented the new supercomputing strategy today
at Super Computing Europe '92. "Work we have already carried out
with leading European academic, research and industrial
organisations has confirmed the viability of IBM's cluster
solutions. The results of some of this collaboration
can be seen here in demonstrations on the IBM stand."
Dr. Budd added: "The European Cluster Initiative I am
announcing today is intended to promote the understanding of
parallel and distributed computing which are major trends in
information technology during the next decade. The IBM Highly
Parallel System is an exciting new development and we look
forward to extending our parallel solutions to satisfy a wider
range of supercomputing requirements."
IBM's parallel systems will utilise AIX*, IBM's implementation
of UNIX, as well as Open Software Foundation standards and the
Institute of Electrical and Electronics Engineers POSIX standard.
The parallel system design will be based on a Multiple Instruction
Multiple Data (MIMD) approach. This will include a distributed
memory architecture, with message passing for processor co-ordination.
(The IBM RISC System/6000 cluster solutions use a similar approach.)
The new systems are designed to take advantage of more than
7,000 AIX applications, already able to run on IBM RISC systems,
and can be used to execute a single job or partitioned for multiple
jobs and users.
IBM will continue to work closely with customers
and application vendors to enhance and parallelise appropriate
scientific and technical applications, to provide the maximum benefit
with IBM's parallel technology.
# # #
* Indicates trademark or registered trademark of the
International Business Machines Corporation.
** Indicates a trademark of UNIX Systems Laboratories, Inc.
Newsgroups: comp.parallel
Path: sparky!uunet!europa.asd.contel.com!gatech!hubcap!fpst
From: steve ("Steve" Stevenson)
Subject: Yet another article on IBM announcement
Message-ID: <1992Feb14.131654.17269@hubcap.clemson.edu>
Sender: fp...@hubcap.clemson.edu (Steve Stevenson)
Organization: Clemson University
Date: Thu, 13 Feb 92 17:01:58 -0500
Approved: para...@hubcap.clemson.edu
Lines: 305
[Again, I found this on the ncsc staff mailing. -steve]
From JE...@WASVMIC1.VNET.IBM.COM Thu Feb 13 14:32:44 1992
Message-Id: <920213193...@duck.ncsc.org>
Received: from WASVMIC1 by vnet.ibm.com (IBM VM SMTP V2R2) with BSMTP id 7928;
Thu, 13 Feb 92 14:32:20 EST
Date: Thu, 13 Feb 92 14:31:23 EST
From: JE...@WASVMIC1.VNET.IBM.COM
To: c...@ncsc.org
Subject: IBM Highly Parallel Qs & As
Status: R
QA's as of 2/11/92, 8:00 p.m. EST
Q1. How does this announcement effect your relationship
with SSI?
A1. Our relationship with SSI remains unchanged.
Q2. Isn't Steve Chen working towards a parallel
supercomputing offering? What differentiates
today's announcement from the SSI project?
A2. Today's announcement complements SSI's efforts.
We are focusing on varying requirements and
customer needs. SSI is developing a high-
high-performance, general-purposesupercomputer
focused on large-scale productionapplications.
Q3. When will products from the SSI relationship be
introduced?
A3. IBM does not make it a practice to speculate on
unannounced products.
Q4. Is there anything new to report on your partnership
with SSI?
A4. No. The research and developmentwork are continuing.
Q5. Rumours indicate that SSI has been seeking additional
financing and/or new partners for their project?
A5. The financial aspects of the partnership are
confidential.
Q6. How does this announcement relate to your agreement with
Thinking Machines Corporation?
A6. When we announced our agreement with TMC, we spoke of
the two companies' intention to be involved in the
cooperative sharing of technology. No joint product
was announced.
Q7. If there are no joint products and no joint marketing
agreements, then you are competing with each other in
the same market, correct?
A7. We have, and will continue to compete with many of the
vendors with whom we have alliances.
Q8. Obviously, TMC is a clear leader in parallel
computing technology. What is IBM planning to do
to compete effectively?
A8. We are very active in the exploration and development
of parallel computing technology and have formed
relationships with other technology leaders, such as TMC.
Both IBM and Thinking Machines Corporation are
dedicated to exploring the many opportunities that
exist in high performance computing -- separately and
together.
Q9. Does this announcement indicate that IBM has decided
highly-parallel systems are "THE technology of the
future" for high performance computing customers?
A9. No. This announcement clarifies that highly parallel computing
is ONE of the computing techniques that we believe will be
important for our customers in the future.
Q10. Doesn't today's RISC-based announcement imply that the
demise of your high-end/mainframe computer
architecture is not far off?
A10. No, not at all.
These technologies are being developed as part of an
overall strategy with elements intended to complement
each other. Each technology relates to the needs and
associated applications requirements for a specific set of
users
The laboratory is dedicated to capitalizing on
all of our resources in order to deliver a scalable
product that will be a standalone highly parallel
system, with the option to integrate to ES/9000
processors.
IBM's large systems provide exceptional facilities
which have been developed over a long period of time.
Many IBM users have found their needs best served when
combined with IBM's large systems strengths
lie. Here are some examples:
- IBM large system solutions offer an integrated and unified
view of the total system as a single image. This provides great
advantage for problems that require very large addressing
spaces, high density of data access, and fine grained
parallelism or microtasking.
- There are a large number of widely used industry programs
that run well on the large system platform without any
modification; vector, scalar, and parallel.
- ES/9000 will remain the system of choice for data intensive
applications, which is one of the reasons for providing the
option to operate the new highly parallel systems closely
integrated to ES/9000 systems, to 'get the best of both worlds'.
- Today's proprietary system software on ES/9000 offers an
exceptional range of system services, such as automatic
backup and recovery of databases, sophisticated security
mechanisms and physical and logical partitioning of the
machine to permit simultaneous multi-mode operation. These
are additional facilities which will make an integrated
traditional large system/highly parallel system solution
attractive to many customers.
Q11. I understand that several IBM business units are
involved in this particular project. What specific
technologies are each of the groups bringing to the
table?
A11. It would be inappropriate to comment on the internal skills
and contributions of each of the associated business units.
However, it is clear that the experience gained with customers
using the RISC System/6000 product from AWD in cluster
configuration plays a significant role in our thinking, as does
the pioneering work which has been underway in parallel computing
within our Research division. Other business units have developed
specific skill bases and expertise, such as the Engineering and
Scientific support centres in the USA and Europe.
Q12. How can someone find out more about IBM's parallel solutions?
A12. Additional information will become available as specific HPSSL
products are launched. The first of these is expected to be
announced later this year.
Q13. When is does IBM expects to be able to provide teraflop
performance?
A13. IBM development efforts depend upon customer demand. The intent
is for the architecture and design of the parallel offerings to
be scalable to the customer's needs. Industry literature would
lead one to believe that there may be a teraflops need in the
second half of the 1990's.
Q14. How does IBM's recent agreement with Bull affect IBM's
parallel processing plans?
A14. It is much too soon to tell.
Q15. Why are 2 IBM centres working on parallel systems in Europe?
A15. In order to satisfy customer requirements in a broad range of
application areas. EPAC (European Petroleum Application Centre)
concentrates on the oil industry. ECSEC (European Centre for
Scientific and Engineering Computing) handles a broad portfolio of
customers ranging from univerisities and research establishments
to industrial companies and banks. Both centres keep each other
updated on their activities, which are complementary. They are also
in close contact with their development colleagues in USA.
Q16. How much does the IBM parallel solution cost?
A16. Specific product prices will be spelled out as products are
announced.
Q17. When will more details be available on the IBM European
Cluster Initiative?
A17. In February 1992.
Q18. What customers are already using IBM cluster solutions?
A18: We are not at liberty to disclose the specifics of our customers'
configurations, except when specifically authorised to do so.
There is, however, significant public literature
on the topic of cluster usage at several US national Labs.
Also, we are in a position to inform you that one of the
early customers for an IBM RISC System/6000 cluster in Europe
is IRSIP (Istituto per la Ricerca sui Sistemi Informatici
Paralleli), Naples in Italy, which is collaborating with
ECSEC on the developement of parallel software tools.
Q19. How many IBM clusters do you plan/expect to install by
year end?
A19. I am unable to divulge IBM confidential business plan
information. I can tell you that a significant number of
customers have expressed interest in the installation of
cluster solutions, as well as the concept of formalizing
those solutions into a packaged offering.
Q20. Why arent you developing the new Highly Parallel System
in Europe?
A20. The Highly Parallel System is a major long-term development
and as such it is appropriate for the two labs with
responsibility for the underlying technology (workstation
and mainframe hardware/system software developement) to
perform most of the work. In this case that means Austin and
Kingston USA respectively. No doubt IBM Europe centres will
play their part, in particular in transmitting European
customer experience and requirements to the US labs.
Q21. Will the new clusters and highly parallel systems be
used for commercial applications?
A21. The product offerings to be produced by the HPSSL will be
specifically oriented toward the wants and needs of the
parallel supercomputing user community. IBM has other efforts
underway to provide world class solutions specifically intended
for the commercial applications set.
Q22. What is the parallel processing technique used by:
(a) the European Cluster Solutions?
(b) the Highly Parallel System under development?
A22. They both use the same initial approach:
- MIMD (Multiple Instruction/Multiple Data)
- Distributed memory architecture
- Message Passing for processor coordination
It is the intent to build on the experiences gained by the cluster
customer set as the parallel offerings are advanced. Thus the
observed requirements of that user community will have
a significant influence on how protocols are developed into the
future.
Q23. IBM recently announced a parallel technology partnership called
InPac. What role does HPSSL play in that program?
A23. The InPac program will be developed and managed within the HPSSL.
InPac is an evolving set of partnerships that will concentrate on
the concept of integration of the parallel environment with the
System 9000 High Performance user. It remains our intent, in
addition to the provision of the product set on an open protocol
basis, to develop and provide the option of close integration with
the System 9000.
Q24. Is the parallel or cluster offering intended to be driven by IBM's
Enhanced Cluster Fortran(ECF)?
A24. The HPSSL products are intended to be implemented on the premise
of a consistent API (Application Program Interface). That means
the standalone version will user FORTRAN consistent with the
industry practices (message passing) and will evolve along with
those platforms toward more powerful parallel capability. The
ECF is an option, in addition to the RISC-based FORTRAN, for
technology experimentation on the System/9000 integrated platform.
At this time, ECF is a part of the InPAC technology program, and
its future as a potential product is dependent upon the join
findings of that program.
Q25. On 9/11/91, IBM made a major commitment to openness. This is
especially important to the engineering and scientific communities.
How does the HPSSL and its direction fit within that commitment?
A25. The products to be developed by the HPSSL will be designed for the
open system environment. The focus will be to provide solutions
with UNIX, TCP/IP, and FORTRAN function. Integration to
ES/9000 will be an added option, to offer specific high-speed
coupling to customers who find it attractive to take advantage of
both technologies.
Q26. Will the parallel products permit the execution of
more than one job at a time?
A26. It is IBM's intent to use the flexibility of the RISC processor to
allow the customer to determine how he chooses to allocate the
processors; some may choose to apply all available power to a
specific application of great computational demand, other may
choose to define the configurations in such a manner as to run
multiple users' problems concurrently.
to determine how he chooses to allocate the processors; some
may choose to apply all available power to a single job
with a heavy computational load, others may choose to operate
the system so as to run many jobs and support many users
concurrently.