| News digest 24 June 2004 |
|
Advertisement
|
 |
| |
| Start |
|
|
| PrimeurLive! from ISC2004 in Heidelberg |
|
This year we again report Live! from Europe's main supercomputing event. See the issues for each day:
This year we start each issue with an experimental "blog" giving you some personal impressions. The next sections provide traditional reporting. The last section "Company news" provides mainly press release type of information from the companies. Hence, a few items are brought to you from three different view points.
|
|
|
|
Advertisement
|
|
| |
| Blog |
|
|
| Germany lost |
|
Yesterday evening was the get-together-party. This means food and drinks in the exhibition area to give the participants the opportunity to talk to the exhibitors. In another part of Europe, Portugal, the European soccer championships are underway. Yesterday evening Germany was playing against the Czech Republic. The Germans had to win to go to the next round. Of course, they lost. This meant that The Netherlands are now progressing to the next round. But there were not enough soccer minded Dutch people at the conference to build a celebration party. Anyway as I was there for the exhibition - and the food and beer, I visited the booths of the University of Manchester, which is participating in an impressive number of Grid computing projects, the NIC at Juelich, showing results of the OpenMolGrid project and a combined booth of SMC networks, Terrascale, Angstrom and some other companies. (AE)
Read further... |  |
|
|
|
| Cray is back |
|
Talked to Jim Rotsolk and Steve Scott from Cray. Seems like Cray is succesful in marketing systems they did not build. The Cray system itself, bought from SGI, and the recent OctigaBay acquisition seem to be received well by the market. The own designed MTA achitecture is not a commercial success. According to Rotsolk, Cray is a relatively small company but very strong in technology. The Octigabay, the Cray offerings and the high-end Red Storm cover the whole spectrum from departmental machines to the TOP3 supercomputers. Steve Scott pointed out, Cray is working on a new architectural platform, targeted for 2006 that should be able to incorporate all the current technologies. It should result in one system image for the user, one file system, one boot and one system management view. In Europe, Cray has recently hired some sales/marketing persons with high-visibility. At Heidelberg I can see they are working hard with (potential) customers. This is an interesting addition to the European market and welcomed by a number of people I talked to. (AE)
|
|
|
|
| Dongarra analyses Tflop/s systems |
|
Jack Dongarra gave an overview of the architectures of the Tflop/s machines that currently exist. Lots of figures on flop/s data rates, performance and dollars. A quick but technical overview. Very good if you want to know a little bit more about current supercomputing architecture. According to Dongarra we are lucky: commodity processors are designed for games, home computing and web servers. We must be grateful that they still include floating point capability. This way you can put computers with hundreds or thousand processors together for a reasonable price. At least for the processor part. Designing a balanced supercomputer for scientific applications, however, is not easy.
Read further... |
|
|
|
| Camp, Weber and Red Storm |
|
Two lectures about the same topic. Different perspectives of course. Sandia is already working on a new architecture for a number of years, since the ASCI Red machine which held the first position in the TOP500 some years ago. All projects ended without a new machine. Red Storm will be different. It is now in testing phase and in September the first row of hardware will be installed, Camp says. Basically Sandia designed the architecture in very detail and then looked for the right partners, that happen to be Cray and AMD. Fred Weber, CTO of AMD then discussed the internals of the machine. In fact, the presentation is more about how difficult it is to produce a commodity processor. Difficult, expensive - a new plant in Dresden costs some 2 billion euro - and highly competitive. Supercomputing in itself is not an interesting market: only 0.1% off all the processors are in supercomputers. Fortunately for us, Weber says, the market has a lighthouse function, which makes it interesting.
|
|
|
|
| Mutter aller Rechner |
|
ISC2004 and especially the TOP500 is a topic in the German press too. The respectable newspaper "Die Zeit" did have a full page on the Earth Simulator, which it calls the "Mother of all computers" (Mutter aller Rechner and the probable next number 1, the Bluegene). Of course, the TOP10 is also listed. In case we would not have time to buy 'Die Zeit", the conference organisers organised a small pile of today's issue, strategically placed near the conference venue's entrance. Earlier this week, the "Computer Zeitung" opened with the new list and, again, IBM is seen as the runner-up: "IBMs Custom-Design jagt Earth Simulator". That does not require translation. This afternoon also featured the vendors' Hot Seat session. As I had the honour to be one of the inquisitors, who were selected to ask (not too) nasty questions, I could not simultaneously write a report on that. This evening is the Castle Party, so further reporting has to wait until tomorrow. (AE).
|
|
|
|
Advertisement
|
|
| |
| TOP500 |
|
|
| Terascale Computing Facility at Virginia Tech to optimize operating environment on system X |
|
Dr. Srinidhi Varadarajan from the Terascale Computing Facility at Virginia Tech held a talk on their experiences with the operating environment on their System X. Since Virginia Tech is a core partner in the upcoming National Lambda Rail network in the United States, the institution has to provide high performance networking capabilities to tie its computational facilities into national computational Grids. Therefore, a Computational Sciences and Engineering (CSE) programme is needed as well as matching computational facilities to complement this CSE programme.
Read further... |
|
|
|
| How will the supercomputer systems and their interconnects of tomorrow differ from their current counterparts? |
|
Jack Dongarra from the University of Tennessee and Oak Ridge National Laboratory examined the processors, the interconnects and the Top131. He distinguished between loosely coupled systems like clusters and tightly coupled ones, such as Cray and NEC vectors and IBM Regattas.
Read further... |
|
|
|
Advertisement
|
|
| |
| Hardware |
|
|
| The world of storage using parallel file systems |
|
Staff computer scientist Peter Haas from the Supercomputer Centre based at the University of Stuttgart (HLRS) started the ISC2004 second day morning session with a detailed overview of the universe of parallel file systems. He sketched the global parallel file system developments by talking about the ASCI PathForward file system strategy; Lustre, Panasas ActiveScale, and IBM SAN FS; and parallel file systems using metadata servers. He addressed the rationale of HSM systems expanding on storage design space, the future of tape, and scalable global parellel HSM systems. He had also something to say on the layering of Legacy Networks, discussing the IEEE 10Gbase-standardization and the backplane and data centre networks.
Read further... |
|
|
|
| Red Storm: what is it and what about the AMD technology |
|
Two speakers discussed the project Red Storm. Bill Camp explained what it is and how it came about. AMD's technology for an evolving HPC world was shown by Fred Weber. Bill Camp outlined the road from ASCI Red to its follower Red Storm.
Read further... |
|
|
|
| |
| Applications |
|
|
| Using Windows as an HPC operating system proves to be a benefit |
|
At ISC 2004 Gerd Heber from the Cornell Theory Center in Ithaca, New York, USA held a talk on high performance computing on Microsoft Windows seen from a user's perspective. He believed that HPC performed on Windows has to be taken more than seriously since Windows and its related technologies are rapidly evolving and the development tools have matured in a considerable way.
Read further... |
|
|
|
| University of Tennessee researchers analyse process fault tolerance on HPC systems |
|
Graham Fagg and Edgar Gabriel won one of the ISC2004 Awards with their paper on the extension of the MPI specification for process fault tolerance on high performance computing systems. They were invited to present their work to the ISC2004 audience and talked about the trends in high end systems with thousand of processors. They tried to define the behaviour of MPI in case an error occurs but stressed that most current systems are robust and do not crash because of a node failure.
Read further... |
|
|
|
| The space simulator is modelling the universe on a budget |
|
For the past several years, a team of University of California astrophysicists working at Los Alamos National Laboratory have been using a cluster of roughly 300 computer processors to model some of the most intriguing aspects of the Universe. Called the Space Simulator, this de facto supercomputer has not only proven itself to be one of the fastest supercomputers in the world, but has also demonstrated that modelling and simulation of complex phenomena, from supernovae to cosmology, can be done on a fairly economical basis.
Read further... |
|
|
|
| |
| Company news |
|
|
| PathScale EKO compiler suite certified as interoperable with Streamline Computing's distributed debugging tool |
|
PathScale Inc. and Streamline Computing Ltd., a UK-based developer of customized clustering solutions, have introduced the mutual certification of their respective Linux compilers and parallel debugger. Specifically, version 1.2 of the PathScale EKO compiler suite and version 1.7 of Streamline's distributed debugging tool (DDT) have been updated, cross-tested and certified to be fully interoperable with each other.
Read further... |
|
|
|
| Breakthrough HP technology yields up to 100 times more bandwidth for Linux clusters |
|
At the Heidelberg supercomputer conference, HP has launched a breakthrough file sharing product that uses new Linux clustering technology to deliver up to 100 times more bandwidth than typical clusters. The new product, HP StorageWorks Scalable File Share (HP SFS), is a self-contained file server that enables bandwidth to be shared by distributing files in parallel across clusters of industry-standard server and storage components.
Read further... |
|
|
|
| More than half of world's Top 500 supercomputers now running on Intel processors (Intel release) |
|
More than half of the 500 fastest supercomputers in the world are now based on Intel Itanium or Xeon processors.
Read further... |
|
|
|
| Voltaire made its debut on the TOP500 list with four supercomputer clusters |
|
Voltaire customers Mississippi State University, Ohio Supercomputer Center, and Sandia National Laboratories all achieved Top500 status with supercomputer-class clusters networked with Voltaire's InfiniBand-based interconnect solutions.
Read further... |
|
|
|
| Dolphin SCI Interconnect Selected for International Space Station Training Simulator |
|
Dolphin Interconnect scalable coherent interface (SCI) adapters have been selected to be used in a very large Raytheon Technical Services Company LLC training system for the international space station. The system uses both the PCI and VME versions of the Dolphin SCI adapters to create a shared memory interface that maps over 300 megabytes of data across clustered PCs and into VME racks.
Read further... |
|
|
