| News digest 19 June 2008 |
|
|
 |
| |
| Start |
|
|
| Primeur Live! from Dresden |
|
Continue over a decade of PrimeurLive! reporting from Eruoep's major supercomputing event, Primeur publishes this three PrimeurLive! issues from ISC08 in Dresden:
|
|
|
|
Advertisement
|
|
| |
| Blog |
|
|
| SUN's Bechtolsheim: It will be difficult to reach Exaflop/s in ten years from now |
|
SUN is back in the top of the HPC market, said Andreas von Bechtelsheim from SUN. They recently installed a 0,5 Pflop/s supercomputer at TACC in Texas. The system was just in time for the TOP500 where it made it into the 5th position. The TACC cluster is providing 60% of the TeraGrid computing power. It took ten years to get from Tflop/s to Pflop/s. But Bechtolsheim sees several fundamental challenges to go to Exascale. He expects these will be hard to solve, so it will be difficult to reach Exaflop/s in ten years from now.
|
|
|
|
| Schedule the schedulers |
|
Moab's Michael Jackson explained a bit about the company's cluster software. It is run on top of other resources schedulers, such as Torque, LSF, and PBS. Moab also claims to help organisations to cut energy costs as much as 50% on idle nodes: it can monitor and schedule temperature and power consumption of nodes. So Moab can take care that heated nodes do not get additional work, but that it is placed on cooler nodes. If it knows nodes will not be used for a few hours, it can put them in power savings mode.
|
|
|
|
| No Roadrunner without Panasas |
|
A Pflop/s computer needs a Petabyte parallel files system. So the Roadrunner uses one: a 3 Pbyte parallel files system from Panasas, said Len Rosenthal from Panasas at the Hot Seat session in Dresden. And not only Roadrunner, but basically any large cluster needs a parallel file system. There is a trend towards a standard for parallel file systems. The standard is called pNFS, or parallel NFS.
|
|
|
|
| Microsoft after the party |
|
Yesterday, the conference party in Dresden was hosted by Microsoft. Movie figures educated us about Microsoft's products and history. For those who did not pay attention, Kyril Faenov did a recap of the HPC developments at Microsoft. The new version of the HPC product is called Compute Cluster 2008 announced earlier this year. The number 23 system in the TOP500 in NCSA runs this product. Microsoft sees something like canonical HPC++ evolving. For instance major computational financing is HPC integrated in the complete analysis process. Grids on Wall Street companies are growing to the tens of thousands of nodes. You cannot program this with MPI. That takes too long, Kyril Faenov says. Hence they created a service oriented application model to support that. There is a Microsoft HPC+ lab for testing and demonstrating these techniques.
|
|
|
|
| QSnet evolution |
|
Duncan Roweth at the Hot Seat session gave a preview of QSnet III. This will be a multi stage network. It is a an evolution of the QSNet design. It is based on increased use of commodity hardware and provides increasing support for the standard software. The system will be 2 x 25 Bit/s links. One of the interesting features is that it includes adaptive routing. This is done as packet by packet dynamic routing. They are building the QSNet into HP Blade clusters.
|
|
|
|
| HP bets on blades |
|
HP's strategy is to provide fast blades, said Frank Baetke in the Hot Seat session. HP has been very succesfull with this during the past year, and even made it into the TOP10 although that was not the main goal. That is providing efficient high-performance computing. For the coming period "scale out" is the message.
|
|
|
|
| Intel to prepare innovative architectures to meet up to future HPC challenges |
|
Intel-based architecture is not just for the large intuitions but is also suited for small businesses and individual developers, according to Richard Dracott during the Hot Seat Session at ISC'08. Although Intel has been busy for some considerable time developing the building blocks - the cores - instead of building full HPC systems, the company is very active at the top edge of the HPC universe. At the ISC'08 Press Conference it was put with a smile by Prof. Dr. Meuer: "There would be no HPC imaginable without Intel."
Read further... |
|
|
|
| Supermicro's SuperBlade is going green |
|
Supermicro's Raphael Wong is concerned about power efficiency as he proved during the Hot Seat Session. Supermicro did a test in their lab and got the award winning SuperBlade. Supermicro have an optimized chassis and are equipped with BIOS and power management tools.
Read further... |
|
|
|
| Cisco to present the FCoE protocol |
|
Dr. Walter Dey from Cisco was concentrating on I/O consolidation and unified fabric. You need a proper amount of slots to cable up the servers. This is becoming a big issue overtime. Instead of having multiple inferfaces you can have only one using I/O consolidation. The new protocol is FCoE.
Read further... |
|
|
|
| Mellanox to highlight Infiniband technology leadership |
|
Michael Kagan from Mellanox talked about Infiniband, the industry standard defined ten years ago, and which is excellent from the price/performance viewpoint. It is reliable with congestion management and very efficient. It is also scalable for petascale computing and beyond. Virtualization acceleration is also possible.
Read further... |
|
|
|
Advertisement
|
|
| |
| TOP500 |
|
|
| Roadrunner - a mini computenik |
|
A few years ago, the Japanese Earth Simulator sky rocketed to the first position in the TOP500 with a performance out of reach for the competition for the years that followed. This triggered a lot of activity in the USA, and also in Europe in the supercomputer arena, because it was felt that they lagged behind. In the US, Jack Dongarra dubbed the term "Computenik" for the Earth Simulator. It led to a lot of activity in planning and developing supercomputers that could bring the USA back on top of the list again: and they succeeded. The new Roadrunner system at position 1 is result of that. The Roadrunner is the first Petaflop/s supercomputer in the world. Is this again a Computenik? Not really, because it was expected more or less. However, we could call it a mini Computenik because now Japan and Europe could say they are lagging behind on this important milestone. To give them the opportunity to start formulating an answer, Hans Meuer organized a special Roadrunner session In Dresden on Wednesday.
Read further... |
|
|
|
Advertisement
|
|
| |
| Hardware |
|
|
| Do not know how to programme multi-core? Use PLASMA |
|
The good thing of multi-core is that you have many CPU's that can do the computational work for you. The bad thing is that you have to tell each of those cores what to do. And you better tell them right, because otherwise most of the cores are just idling: eating up electricity without doing anything useful. So you have to program a multicore system. The more cores the more difficult it becomes. There is also a big difference in approach between different types of multi-core. For instance a multi-core CELL processor is quit different from a standard Intel multi-core processor. And where do you start? Program from 4 cores and optimize that to 8 cores in two years time? If you need matrix operations a lot, you do not have to worry about multicore. Jack Dongarra and his team is taking care of that. They are working on a library called PLASMA that will do all the matrix operations in an optimized version for you. But you have to wait some time: PLASMA is still under development.
Read further... |
|
|
|
| Future HPC applications in need of multicore and manycore platforms |
|
Justine Rattner from Intel, USA elaborated on the exascale visions Intel dreams about to bring supercomputing to the everyday user. He made a comparison between the dataset size required in the applications including medicine, personal media, travel and learning, and entertainment versus the required performance. Which types of technology are needed to respond to the user and application needs? Intel is convinced that there is a whole new potential of emerging killer apps that advanced HPC technology will help make come true.
Read further... |
|
|
|
| |
| The Grid |
|
|
| Partnership for Advanced Computing in Europe rides the wave of ambition |
|
In the morning session Prof. Dr. Achim Bachem provided an overview of the PRACE plans to develop an HPC Research Infrastructure in Europe that also includes the industry. PRACE stands for Partnership for Advanced Computing in Europe. HPC is a key technology, according to the speaker. Supercomputers are the tool for solving the most challenging problems through simulations. The access to capability computers of leadership class is essential for international competitiveness in science and engineering. Providing competitive HPC is a must for Europe.
Read further... |
|
|
|
| ParMA project team draws the multi-core card in their parallel programming business |
|
The goal of the Bird-of-a-Feather Session featuring the ParMA project team was to present early results of the ITEA2 ParMA project, which stands for Parallel Programming for Multi-core Architectures. The project has been launched by European HPC players from France, Germany, Spain, and the United Kingdom to help the HPC community to take full advantage of the power of multi-core architectures. The 17 partners leverage their professional and scientific competence to evolve design and programming models to help develop and restructure parallel applications; extend and integrate tools for performance analysis and debugging of parallel applications; develop, restructure, and optimize parallel applications from diverse domains, e.g. simulation, automotive, avionics, etc., for multi-core architectures; and explore parallelization strategies to optimize the performance of the generated code. The presenters showed how well some code patterns can be improved.
Read further... |
|
|
|
| |
| Company news |
|
|
| Sun breaks into top five on TOP500 Supercomputers List with highest ranking open HPC system |
|
The first Sun Constellation System-powered supercomputer at the Texas Advanced Computing Center (TACC) captured the number four spot on the TOP500 Supercomputers list - the highest ranking system based on an open architecture. Demonstrating the pervasiveness of Sun technology on the list, Sun HPC software and storage also made strong showings on the list with Sun's Lustre file system managing data on six of the list's top 10 sites as well as nearly half of the top 50 supercomputers. Additionally, half of the list's top 10 sites and nearly half of the top 50 systems archive their data on Sun storage.
Read further... |
|
|
|
| Voltaire powers world's most powerful supercomputer for NNSA's Los Alamos |
|
Voltaire Ltd.'s InfiniBand-based switches are powering the world's largest supercomputer for the National Nuclear Security Administration's (NNSA) Los Alamos National Laboratory called Roadrunner.
Read further... |
|
|
|
| Mitrionics and Nallatech announce PCI Express FPGA Accelerator Kit targeting defense and bio industries |
|
Mitrionic and Nallatech have made available an FPGA Accelerator Kit targeted at customers within the defense and bioinformatics sectors interested in leveraging the benefits of Commercial Off The Shelf (COTS) products. The kit comprises of an 8-lane PCI Express card featuring a Xilinx FPGA directly coupled to multiple independent banks of SDRAM and SRAM memory.
Read further... |
|
|
|
| HP BladeSystem servers occupy 35 percent of TOP500 List of world's most powerful supercomputers |
|
HP continues to strengthen its position on the TOP500 list, a ranking of the world's 500 most powerful supercomputers, with HP BladeSystem c-Class servers dominating as the most prominent computing architecture on the list. HP BladeSystem c-Class servers power 176 entries, or 35 percent of the total entries on the TOP500 list. This is more than any other single computing architecture and more blade installations than all other vendors combined. Blade systems have experienced strong momentum in high-performance computing, with 66 percent of the supercomputers on the TOP500 list, 329 total entries, now configured with blade servers.
Read further... |
|
|
|
| Sun expands Sun blade family with new four socket blade server for HPC and enterprise applications |
|
Sun Microsystems has made available its newest Sun Blade x64 system, offering industry-leading memory capacity to run the most compute-intensive HPC and enterprise applications. The Sun Blade X6450 Server Moduleis powered by four high-performance dual-or-quad core Intel Xeon processor 7300 series. It can run several operating systems, including the Solaris 10 Operating System (OS), Linux, Windows and VMware, the Sun Blade X6450 Server Module gives customers the flexibility to run existing 32-bit applications as they migrate to 64-bit applications.
Read further... |
|
|
|
| QLogic announces general availability of world's fastest InfiniBand HCA, based on QLogic TrueScale ASIC Platform |
|
QLogic Corp., an expert in networking for storage and high performance computing (HPC), is now shipping in volume the highest performing InfiniBand host channel adapter (HCA) in the industry, based on the QLogic TrueScale ASIC platform for Double Data Rate (DDR) and Quad Data Rate (QDR) InfiniBand.
Read further... |
|
|
|
| Voltaire announces 40 Gbps InfiniBand Switch development plans |
|
Voltaire plans to develop a 40 gigabits per second (Gbps)/quad data rate (QDR) switching platform based on the new Mellanox InfiniBand InfiniScale IV switch silicon announced last week. The new silicon enables the design of 40 Gbps InfiniBand switches with higher densities, improved scalability and lower latency than 10 or 20 Gbps InfiniBand switch offerings.
Read further... |
|
|
|
| Windows HPC server debuts in Top 25 of world's TOP500 largest supercomputers |
|
Microsoft Corp. debuted in the top 25 of the world's top 500 largest supercomputers with the National Center for Supercomputing Applications (NCSA), which ranked at no. 23 with 68.5 teraflops. The company also released the candidate version of Windows HPC Server 2008 will be available for download in the last week of June.
Read further... |
|
|
