2008/9/16 Greg Lindahl wrote:
Tuesday, 4 November 2008
Beowulf Exchange
2008/9/16 Greg Lindahl wrote:
Monday, 3 November 2008
Useful Links for Quadrics
Quadrics Website
http://quadrics.dphoto.com/
Quadrics PhotoGallery
http://web1.quadrics.com/twiki/bin/view/Main/WebHome
Quadrics FAQ wiki
http://quadrics.bttradespace.com/
Quadrics space on BT Tradespace
http://www.top500.org/
Top500 website
Promotion of Supercomputers and Petacomputing Technology (PROSPECT)
Quadrics has joined alliances that promote technical developments and partnerships that promote HPC technology both within Europe and across the world.
PROSPECT is a European HPC alliance formed in 2007 and joined by companies like Intel, IBM, Quadrics and ParTec, HPC facilities like Barcelona Supercomputing Centre, Cambridge Scientific Computing Centre, the Deutsche Wetterdienst, Jülich Supercomputing Centre, the Leibniz Rechenzentrum and several universities. PROSPECT's mission is to closely collaborate with the 'Partnership for Advanced Computing in Europe' (PRACE) initiative to establish a European HPC technology platform, to define research and development priorities, time frames and action plans in supercomputing for the benefit of Europe's growth, competitiveness and sustainability. Quite a few development projects have been created by the PROSPECT partners with the goal to advance technologies for the multi-Petaflop/s era.
Friday, 31 October 2008
Quadrics Partnerships and Alliances: CDA
Leveraging the ControlPoint Software Suite, which is available as source code to members of the Alliance, equipment providers and ODMs can deliver unique system solutions quickly and cost effectively, and researchers can leverage ControlPoint and Fulcrum's full-featured FocalPoint reference designs to experiment with new protocols and network topologies. Quadrics is one of the founding members together with Fulcrum Microsystems, Continuous Computing, Green Hills Software, Liquid Computing, Nimbus Data Systems, Open Grid Computing, Panasas, SMC Networks and XORP.
Quadrics announces new everyday low pricing on Quadrics TG201, 24-port 10 Gigabit Ethernet Switch
Wednesday, 29 October 2008
10 GigE market is already sizeable - amounting to $9.5 billion this year alone - and is poised for impressive growth going forward.
10 Gigabit Ethernet Going Mainstream
"A majority of service providers we've spoken to are expecting to invest in 40G until the 100G market is up and running; some providers are hoping to skip the 40G phase altogether, but we don't see that being a viable option, as growing traffic demands are outstripping current capacities and 100G won't reach reasonable price points until about 2012 or 2013," said Principal Analyst and Infonetics Co-Founder Michael Howard in a statement. "When 100G Ethernet arrives, it'll be the next big thing and the most important, because it will last to at least 2025, solving traffic problems for a very long time."
What's driving 10 GigE -- and, for that matter 40 GigE -- adoption? It's the capacity, stupid.
"As corporate applications get more and more bandwidth-intensive and enterprises centralize their resources, a tremendous strain is being put on their networks," noted Matthias Machowinski, directing analyst for enterprise voice and data at Infonetics Research, in a release. According to Machowinski, network managers cite a need for increased capacity as their No. 1 overall initiative.
"[I]t's no surprise...that shipments of 10 [GigE] ports are going through the roof," Machowinski said. "Ethernet switch-based 10 [GigE] port shipments, for example, grew 140 percent in 2007. Growth should remain healthy for years to come, as this is an area where buyers are putting their money where their mouths are."
Over the next few years, Infonetics projected, 40 GigE revenues will grow at a healthy 59 percent annual clip. In 2007, it said, vendors shipped slightly more than 1 million 10 GigE, 40 GigE, and 100 GigE ports on enterprise- and service provider-grade gear; by 2011, Infonetics suggested, that total will reach 7.4 mllion. --
Monday, 27 October 2008
Quadrics' Scalable 10 GigE Solutions
Advantages of 10GbE
10 Gigabit Ethernet is the natural evolution of the well-established IEEE 802.3 standard in speed and distance. It extends Ethernet's proven value set and economics to metropolitan and wide area networks by providing:
- Potentially lowest total cost of ownership (infrastructure/operational/human capital)
- Straightforward migration to higher performance levels
- Proven multi-vendor and installed base interoperability (Plug and Play)
- Familiar network management feature set
- Supports all traffic types-data, voice and video over IP
- 10 Gigabit Ethernet does not obsolete current investments in network infrastructure
- Flexibility in network design
- Ability to support jumbo or large frame sizes to reduce the network overhead
The price-performance that the combination of Gigabit and 10G bring to customers is driving increasing adoption of 10G. In general as desktops become more powerful and applications more feature-rich, it generates a lot of data/content, which necessitates a more powerful transport infrastructure of which Gigabit and 10G networks are a prime example.
Some of the applications driving this technology are Web 2.0 applications typically with multimedia content. There are specific applications in banking and finance as well as CAD applications in the manufacturing sector. Media companies using animation and high-end graphics applications are the early adopters of 10G technology in the industry.
Chandra Kopparapu, Vice President for Asia Pacific at Foundry Networks, said, “10GbE enables the use of Ethernet for transporting data, voice and video traffic within an enterprise and a carrier’s network. With 10G, companies will have the capability to provide GbE service to workstations and, eventually, to the desktop in order to support applications such as streaming video, medical imaging, centralized applications, and high-end graphics. New and upcoming bandwidth-hungry applications like streaming video, medical imaging, centralized applications, and high-end graphics will drive 10GbE technology. Verticals like entertainment and animation have shown great interest in 10G.”
Extract from: Gigabit Ethernet and beyond
Thursday, 23 October 2008
“The New Secret Weapon.”
Wednesday, 22 October 2008
Bull lands 200 teraflop German super deal
The cluster at the German HPC lab will be a mix of Bull NovaScale rack-mounted servers, and X64 blade servers manufactured by Sun Microsystems. The Novascale servers are based on Intel's Xeon X64 processors and support Linux, Windows, and Bull's GCOS mainframe environment.
The Juropa project - short for Jülich Research on Petaflops Architectures - chose its prototype cluster platform in June 2007, based on IBM System x servers using Intel's dual-core "Woodcrest" Xeons. This is an interconnect made by Quadrics (part of Italian conglomerate Finmeccanica). The project also includes cluster management software from ParTec, which is located in Munich.
This time around, the mix of Bull and Sun iron will be based on Intel's forthcoming eight-core "Nehalem" Xeon processors - IBM did not get the hardware deal. Bull did not say what interconnect it has chosen for the kicker Juropa cluster, but the prototype used a variant of the Quadrics QsNet-II protocol that rides atop 10 Gigabit Ethernet links called QsTenG. The organization said that the future production system would be based on the QsNet-III protocol, which can scale up to 16,384 nodes.
Sun is kicking in its Zettabyte File System for Solaris 10 and Linux, in addition to supplying the Nehalem blade servers. It also last September picked up the Lustre cluster file system, an open source file system for HPC clusters.
Forschungszentrum Jülich says that the new iron, installed in early 2009, will boost performance on the Juropa machine by a factor of 20. The center is also a big user of IBM's BlueGene massively-parallel PowerPC-Linux supercomputer, and currently has number six on the Top 500 list with a 65,536 core monster rated at a sustained performance of 222.8 teraflops.
From: http://www.channelregister.co.uk/2008/10/06/bull_german_super/
Monday, 20 October 2008
Write up a case study to get up to USD 4,000 discount off your next purchased TG201!
Price List of Quadrics' 24 ports 10 Gigabit Ethernet Switches
TG201-CA 24 Port 10GbE CX4 ports managed switch | $14,400 |
TG201-XA 24 Port 10GbE, 12 CX4 and 12 XFP managed switch | $16,200 |
Optical transceiver modules QS-QM688-LR Module XFP Transceiver LR (1300nm) | $900 |
QS-QM688-SR Module XFP Transceiver SR (850nm) | $500 |
Saturday, 18 October 2008
Quadrics photogallery online
To access it either click on the image on the right, or bookmark http://quadrics.dphoto.com
Friday, 17 October 2008
QsTenG 24-port 10 Gigabit Ethernet Switch
Industry's lowest latency - 10Gb Ethernet Switch with flexible interfaces, CX4 and XFP LR/SRThe Quadrics TG201, 24-port 10 Gigabit Ethernet Switch leads the market in ultra low-latency intelligent switching, delivering cut-through latency as low as 200ns. With its robust layer-2 switching capabilities and the ubiquity of Ethernet, the Quadrics 10 Gigabit Ethernet TG201 switch is uniquely placed to service the ever-demanding needs of the Data Centres, High utilization virtualized servers, Campus network core switching and storage consolidation.
QsTenG TG201 Features
- 480Gbps of throughput
- Fully non-blocking layer-2 switching
- Industry's lowest latency - 200ns
- Option to use fibre with XFP pluggable modules enabling LR/SR media
- Option to lower TCO by using CX4 ports for in-rack connectivity
Article on Cluster Interconnects from HPCCommunity Kusu toolkit
3.3 Cluster Interconnects.
Most small clusters have only one network and this is typically Gigabit Ethernet.
For larger clusters it is common to see i) Administration network typically GE, ii) HPC Network usually one of Infiniband, Quadrics or Myrinet, iii) Out-of-Band network.
The focus of this section is on the HPC Network.
Firstly determine if you your code will benefit from a HPC network. For bandwidth sensitive codes or IO intensive codes, then using Infiniband, Quadrics or Myrinet would help tremendously.
Which HPC network is the best? For that answer - you would have to know the characteristics of your code. Best to run your applications on the network you intend to buy - beg and borrow resources from your friends and colleagues in the community or the vendor to test.
In real-world MPI codes, Quadrics is typically the fastest due to its SHMEM capabilities. Quadrics also has excellent MPI management software in the form of RMS.
For people in the know, they are willing to pay the premium for Quadrics.
Read in full on this link.
Thursday, 16 October 2008
Ethernet or Ether-not
Wednesday, 15 October 2008
The new supercomputer: Albert3
The new BMW Sauber supercomputer, Albert3, again with Quadrics interconnect, will have 4,224 processor cores and 50.7 Teraflops computing power.For more information please visit: bmw-sauber-f1 .
The BMW Sauber F1 Team presented its supercomputer Albert2 in December 2006. The 21-tonne facility for CFD calculations was already based on Intel technology (consisting of processors, the motherboard, chip set and server housing) and was at the time of start-up one of the most powerful in Formula One. Albert2 had 256 nodes with two Intel®Xeon® 5160 Dual Core processors each (two cores per processor). That amounted to a total of 1,024 processor cores. The capacity of the main memory was 2,048 GB and maximum computing power was 12.28 TFlops (12,288 GFlops). It wasn’t long before another 32 nodes were added, bringing the total to 288 nodes or 1,152 processor cores. Now the BMW Sauber F1 Team has ushered in the next stage and upgraded the existing computer. 384 nodes equipped with Intel®Xeon®E5472 Quad Core processors (four cores per processor) and related Intel technology have been added to the previous system, which means the new Albert3 supercomputer now has a total of more than 4,224 processor cores at its disposal. The computer’s RAM has grown to 8,448 GB and maximum computing power to 50.7 TFlops. That is a phenomenal 50,700,000,000,000 computations per second. To match this computing performance, the entire population of Munich and Berlin (4.7 million) would have to multiply two eightdigit numbers every three seconds for a whole year. In the current top 500 rankings of active supercomputers worldwide, this performance merits 45th place (third place among systems in industrial use). The new supercomputer, developed like its forerunner by the Swiss company Dalco and running on CFD software by Ansys-Fluent, weighs in at 38 tonnes yet has a footprint of just 24 square metres. The tremendous technical potential of Albert3 is harnessed for analysis in the field of aerodynamics. With its help, the specialists calculate components for the Formula One race cars, using grid models often made up of more than 100 million cells. CFD plays a particularly crucial role in the development of front, rear and auxiliary wings, as well as in engine and brake cooling. Computer-aided air flow simulation is not in competition with the work in the wind tunnel, but rather complements it. “A great advantage of CFD is that you can represent the air stream visually and that way understand why one component is better than another,” explains Willem Toet, Head of Aerodynamics. BMW Motorsport Director Mario Theissen says: “Unlike the other teams, we are not planning to build a second Formula One wind tunnel, but in future will continue to concentrate on the steadily expanding possibilities in the field of simulation.”
Acknowledgements: this article appeared on automobilesreview website on August 6th 2008.
Tuesday, 14 October 2008
BMW Sauber Albert2 Supercomputer speaks!
Acknowledgements: This conversation with the BMW Sauber F1 Team Albert2 took place in September 2007.
The BMW Sauber F1 Team has a worker of superhuman diligence within its ranks. His name? Albert2. Based in Hinwil and powered by Intel technology, this supercomputer makes a significant contribution to the team's success on the race track. Albert2 is a mathematical genius of literally huge proportions. But, as he tells us, a day off does not figure in his calculations.
Albert2: Ever since my father Albert took his well-earned retirement in December 2006. So you could say that the connection with motor sport runs in our family.
Question: How would you define your role within the team?
Albert2: As I work around five times more quickly than my father, my boss - our Technical Director Willy Rampf - has given me some important jobs to do. My main strengths lie in the field of Computational Fluid Dynamics (CFD), which is extremely important in the development of front, rear and auxiliary wings, for example. Before an idea from the engineers is tested in the wind tunnel, I simulate the effects of a new component on the car's performance. And that saves the team a lot of time.
Question: If you do such an important job, why do you sit - almost hidden away - in the basement of the wind tunnel building in Hinwil?
Albert2: There is a very simple reason why. When you weigh 21 tonnes and measure ten metres in length, things would get rather tight in a normal office. Besides, it's not where you're sitting that matters, but what you can do for the team.
Question: How closely do you work with your colleagues?
Albert2: Formula One is a team sport, in which you have to work hand-in-hand with one another. I link up with the wind tunnel, for example, at many different stages of the development process. Because we spend our time concentrating on similar tasks, some people think that we are actually working against each other. But that is far from the case. On the contrary, we support each other's work. The wind tunnel tells everybody if I was right with my estimates. At the end of the day, if Nick Heidfeld or Robert Kubica improve their speed then we will both be smiling, so to speak.
Question: Are you a keen sportsman yourself?
Albert2: Being a man of few words and with my, shall we say, statuesque physique, I can come across as rather ponderous at first - but I'm quite a sprightly chap really. My 512 Intel Xeon 5160 processors allow me to run 12,288,000,000,000 calculations per second. To put that into perspective, every one of the 1.3 million inhabitants of Munich would have to multiply two eight-digit numbers every three and a half seconds over the course of a whole year to generate the same performance. So to answer your question, I'm rather better at mental exercise than physical sport.
Question: Can you give us some concrete examples of what you do?
Albert2: In the wind tunnel, for instance, there is practically no way you can simulate how a tyre deforms through a corner. It is pretty much impossible to generate these heavy loads artificially. By contrast, I work with a model of the BMW Sauber F1.07 consisting of 100 million cells, all of which interact with one another. This kind of simulation is a straightforward job for me. I exert pressure on the tyres and highlight the areas of the car affected. In this way, you can see fairly quickly what kind of effect the tyre deformation through corners will have.
Question: Do you keep an eye on what the team's rivals' computers are up to?
Albert2: Of course, that's all part of Formula One. But I have no reason to worry. A supercomputer was recently unveiled in England, but it is much slower than me. It can only process eight teraflops - i.e. eight trillion floating point operations per second. As I said earlier, I can get through almost 12.3 teraflops.
Question: Do you ever take any time off?
Albert2: No. For a supercomputer there is no such thing as controlled working hours and we don't have any workers' representatives to look after us. I work day and night to bring about improvements, and that's the way I like it. For me, being shut down would be the biggest insult. I wouldn't know what to do with myself. And anyway, who would take me away on holiday with them? I'm hardly a slimline laptop, after all!
Quadrics machines: Albert2 for BMW Sauber - Automotive
The compute power will be used primarily for Computational Fluid Dynamics (CFD) calculations, using the ANSYS-Fluent software package, which is becoming increasingly important to compete for split-second advantages per race lap.
The Supercomputer was designed and built by DALCO using latest Intel Xeon technology to achieve its tremendous power. Thanks to the sophisticated design and the selection of the proper components "Albert2" is 5.5 times faster than its predecessor that is only two years old.
The compute section of the cluster consists of 512 Dual Core Intel Xeon 5160 processors (1024 processor cores), each running at 3.0GHz and with 4 Megabytes cache, 2,048 Gigabytes of physical memory and more than 20'480 Gigabytes of dedicated disk storage. These components are integrated with the Intel motherboard which uses sophisticated snoop filters for enhanced performance. Each compute node uses the Intel SR1500 chassis with optimized thermal characteristics. Two fail-safe Master units drive and supervise the computing complex.
"It is amazing how much performance we were able to get with the Intel Xeon technology in combination with the Fluent CFD software", stated Christian Dallmann, founder and CEO of DALCO. "My team can be very proud of having achieved the highest level of efficiency, which amazed experts." Francois Dallmann, lead engineer at DALCO, added: "These outstanding results are possible because of our extensive know-how in this area and the distinctive selection of top-quality components, combined with Swiss quality work."
Quadrics' ultra-low latency high-bandwidth interconnect which provides high, sustained performance for the complex data exchange among the processor cores was selected for this critical task because extensive benchmarks proved that it offered the best overall data rates and lowest latency.
All data files are stored on a Parallel Cluster file system. The demand for massive, parallel data movement is being met by the system's scalable performance and capacity based on an object-based approach to storage.
Data movement, job processing and cluster management is performed via a no blocking high performance Gigabit switch which provides full network bandwidth to every single compute node.
The cluster utilizes SUSE 64-bit open-source Linux, integrated and optimized using DALCO's own cluster management software including enhanced system and environmental monitoring and best practices.
The Supercomputer is integrated into ten water cooled racks from APC with an overall dimension of 10 metres wide, 1.2 metres depth and 2.3 metres height and total weight of 21 tons. The total power consumption under full load is 122 KW. Compared to its forerunner the new Supercomputer provides nearly 4 times the performance per Watt.
"Albert2" is capable of performing 12,288,000,000,000 calculations per second. To achieve the same computing performance, the entire population of Munich (1.3 million) would have to multiply two eight-digit numbers every three and a half seconds for roughly a year.
Announced in August 2008, the new BMW Sauber supercomputer, Albert3, again with Quadrics interconnect, will have 4,224 processor cores and 50.7 TFlops computing power.For more information please visit www.quadrics.com
Monday, 13 October 2008
What is HPC?
Acknowledgements: from sharcnet website.
Sunday, 12 October 2008
Meet Quadrics at SC08
If you are interested in meeting Quadrics' technical experts and business development managers, please fill in your details below.
Meeting point, unless you specify otherwise in the form, will be the Quadrics stand, number 313.
Book a meeting with Quadrics at SC08
QsNet3 or QsNetIII?
Saturday, 11 October 2008
Quadrics, leading developer and supplier of interconnect technology for HPC
Quadrics products form an integral part of some of the world's largest computers, allowing many thousands of processors to be used as a single system, and have been selected by the world's leading vendors and integrators as well as major supercomputing centres.
Quadrics' third generation of interconnect product, QsNet3, is a multi-stage or 'fat tree' switch network connecting commodity servers. It has been specifically designed to maximise the performance of large HPC applications.
QsNet3 comprises adapters, switches, cables allowing the construction of federated network of up to 4096 nodes. QsNet3 software includes support for HPC programming tools including MPI, Shmem, UPC and Lustre. HPC specific optimisations include low latency, high bandwidth links, adaptive routing (used extensively to distribute network load), hardware barrier and broadcast support. Quadrics interconnect is particularly suited to large simulation programs that require the combination of outstanding network performance and 24 7 production execution.
For information about Quadrics please see our website, www.quadrics.com.
Quadrics at SC08
At the show, Quadrics will be launching its third generation of interconnect product. QsNetIII is a multi-stage or 'fat tree' switch network connecting commodity servers. It has been specifically designed to maximise the performance of large HPC applications.
If you want to come and see why not popping in at stand 313? Alternatively, go to the Quadrics website, www.quadrics.com, where you can book a meeting.