Features:
IDC HPC USER FORUM HELD IN EUROPE
by Christopher Lazou
Over 40 experts attended the one day IDC user forum held at the IBM offices
in central London with the aim of promoting the HPC industry and share
experiences within this eclectic community. The usual format was adopted with
various vendor and user presentations. The vendors peddling their products
were IBM, SGI, Sun Microsystems and Streamline Computing, a small UK company
specialising in building cluster systems. The day was crammed with a total of
14 presentations and with only half an hour for lunch.
Examples of speakers describing the European user scene, included Professor
Murat Kunt, Swiss Federal Institute, Lausanne, Anwan Osseyran, SARA, The
Netherland, using SGI and IBM systems, Dr. wolfgang Sell, Deutsches
Klimarechenzentrum GmbH (DKRZ), Germany, using NEC SX-6 systems, Phillip
Morris, White Rose Computational Grid, UK, using Sun machines and Daryl
Landeg, AWE (Atomic Weapons Establishment), UK, using IBM P3 systems.
From the USA, Allan Snavely, from UCSD described Benchmark activities, Larry
Davis, DOD, gave a talk on computational chemistry and a short update on the
activities of the High-End Computing Revitalization Task Force (HECRTF) (see
HPCwire 09.26.03). Paul Muzio, from the Army High Performance Computing
Research Centre and Suresh Shukla, from Boeing, provided insights from their
experiences and results from running large applications on their computer
facilities. They both expressed relief and delight that after several years in
the wilderness, USA computer users are again allowed to buy vector systems,
namely, the US produced Cray X1. IDC then wrapped up describing some recent
IDC market results. Below are a few extracts mostly from European users to
give a flavour of the meeting.
Professor Murat Kunt, described his research in signal processing. He stated
that digital imaging in medicine and other fields generate large amounts of
data. For example, one hour of colour video conferencing generates a terabit
of data. Many problems need addressing not least a better understanding of
human visual system and finding better data compression techniques. His
research has instigated several adaptations to conventional practices,
abandoning orthogonal and canonical forms. He is using data content adaptive
sampling and multiple methods for information processing for tackling a single
problem with amazingly efficient high quality results. The examples he gave
looked convincing.
Dr. Anwar Osseyran, director of SARA, described their computer facilities and
services provided to the Netherlands academia and industry. Their hardware
include 1024 processors of SGI 3800, 416 processors SGI Altix 3700 (2.4Tflop/s
peak) an IBM cluster 1600 with about half teraflop/s peak performance and
several Linux clusters. They also have a CAVETM (immersing virtual reality
environment) for visualisation services. He duelled on the challenges of
extracting maximum return from hardware investment, higher throughput, running
larger user jobs, doing better simulations and improving efficiency by
optimising single CPU utilization. Data growth is doubling every year and data
management is becoming a big challenge. The current capacity of 200Terabytes
is expected to rise to 1.2Petabytes in the next three years. They are
considering an upgrade to a storage system allowing the share of all storage
resources online. They are also participating in the EU Data Grid for sharing
data files between computer facilities throughout Europe. This is of
particular interest to life sciences, for bio-informatics and genomic data,
which can then be visualised using the CAVETM facility.
Dr. Wolfgang Sell, director of the Deutsches Klimarechenzentrum GmbH (DKRZ),
Hamburg, gave a brief description of their activities. DKRZ was founded in
1987 with the mission to provide state-of-the-art supercomputing, data
handling and associated services, including high level visualization, to the
German scientific community, to conduct large scale earth system and climate
modelling.
DKRZ recently upgraded its computer systems and became one of Europe's fastest
supercomputer facilities in production, used for climate research. They now
have the latest supercomputers from NEC, the SX-6 series, a unified data
management system, based on the Intel IA-64 (Itanium) architecture and Linux.
These elements are the core of the modernization project costing Euro 34
million.
To summarise briefly, the final phase required delivery of at least
0.5Tflop/s, sustained performance, at least 5GBytes/s of data rates,
transparent data access and management for HSM, around 50TeraBytes of disk
cache and 1.4PetaBytes of tape drives.
After a competitive procurement exercise the contract was awarded to NEC High
Performance Computing Europe (HPC Europe), the only vendor offering the
compute power and functionality to satisfy DKRZ's needs, as well as the system
integration for a total solution. NEC HPC Europe offered a 192CPUs vector
parallel NEC SX-6 system with a main memory of 1.5Terabytes and 1.5Teraflop/s
peak performance for the high performance compute server and the AsAmA series
- TX7 system with Intel Itanium (Madison) processors - for the data-handling
server.
NEC HPC Europe successfully completed the installation of the DKRZ system on
time. To achieve this, NEC used its own products, but also incorporated
elements from other hardware and software vendors, StorageTek, the Legato
hierarchical file system (GFS) and the ORACLE database running on top of
Linux, to deliver an optimal solution.
As Wolfgang Sell, said: "We are very happy with NEC HPC Europe. They looked
for the most appropriate state-of-the-art technologies available in the market
today, both NEC products and best system components from third party vendors
and integrated them to deliver the most advanced solution for our user
applications. This was done within the original budget, in a seamless non-
disruptive fashion. The architecture adopted is scalable to allow expansion
and open enough for the introduction of new hardware and software for at least
the next decade."
Darel Landeg, computer director at AWE, gave a history of their computer
facility from the first Cray 1, to the last upgrade in 2002 when they
installed an IBM SP (Blue Oak) with 2.88Tflop/s peak performance. AWE is run
by the MOD in the UK and has roughly the combine role of Los Alamos and Sandia
Laboratories. Its mission is to underwrite the safety and performance of
Trident and develop the capability to produce a successor system without
recourse to underground nuclear warhead tests. They are relying on laboratory
tests of simulated components and eventually they plan to migrate to a model
based system, relying wholly on a virtual representation of a weapon system.
They expect to take a decade to put the infrastructure in place including
software tools. The HPC infrastructure provides the data backbone and
simulation capability for this endeavour. Darel went on to describe some
hydrodynamic (CFD) codes used to simulate turbulent mixing using various
techniques and 1Billion cells. He estimated that their work would require
about 25Teraflop/s by year 2005 and hundreds of Teraflop/s by year 2010. More
importantly one must adopt a balanced development for supporting computing
infrastructure, compute power, network, storage and desktop visualization
systems. AWE also adopted a dual data centre approach for resilience.
Paul Muzio from the Army High Performance Computing Research Centre, USA,
described their new Cray X1, procured as part of the DOD modernization
programme. He gave some very positive performance results obtained using the
Cray X1. For example, on a Finite Element Method CFD code, the Cray X1
achieved a speedup of 32 times compared to the Cray T3E and a speedup of about
14 times compared to the IBM p690 P4. He stated that the Cray X1 is suited to
large problems. It is about 4 times faster than the IBM P4 on small problems
and about 25 times faster on large problems. This was illustrated using the
NCAR MM5 weather forecasting model. For a forecast of the whole of the USA,
using a 5Km grid, this larger problem showed the Cray X1 performing much
better. In general Paul found that the Cray X1 delivers good performance on
large applications, ranging from 18% to 33% efficiency.
Suresh Shukla, computer manager at Boeing described briefly their computing
facilities. These include, CrayT916/T32 systems, SGI Origin 3800 and IBM SP
(Regatta). He also announced that they ordered a 64 MSP Cray X1 to be
delivered in December and expected to become operational by February 2004.
Suresh then discussed decision-making and how to justify computer costs to
higher management. The key is to match user inputs to requirements, de-
emphasize cost of hardware and highlight total cost/benefit to the
organization. For example, emphasize the benefits of improved turnaround for
designing new products He concluded by saying: "By solving yesterday's
problems we are stunting science, damaging future competitiveness. Currently
Boeing does not have a supercomputer. A supercomputer should have fastest
single CPU, be good on scalar, good on vector and good on parallel codes. The
Cray X1 and the NEC SX-6 are the only supercomputers today".
Finally, the IDC HPC market trends analysis provided a mix bag of findings.
The bad news is that the overall HPC revenue was down by 7.2% to $4.7Billion
in year 2002 compared to year 2001. The good news is that revenue from
capability computing was up by 24.3% to $1Billion. Overall it will take 5
years for HPC market revenue to climb to year 2000 level. Technical servers
are expected to provide $6.3Billion by year 2007 of which about $3Billion
would be from capability computing. Traditional HPC is likely to grow modestly
by 1.4%, while bioscience is expected to grow by 16.6% per year. The number of
systems and clusters are expected to grow dramatically.
(Brands and names are the property of their respective owners) Copyright:
Christopher Lazou, HiPerCom Consultants, Ltd., UK. Email:
Chris@lazou.demon.co.uk, October 2003.
The opinions expressed in this feature are solely those of the author and do
not necessarily reflect those of HPCwire, its publisher or its staff.
|
