Editor’s note: We have reposted this video with a much better audio track.
In this video, Matt Leininger from Lawrence Livermore National Laboratory presents on LLNL’s partnership with QLogic and Dell to enable a new level of performance for commodity Linux clusters.
Recorded at the QLogic booth at SC10 in New Orleans on Nov. 17, 2010.
One of the cooler things I saw at SC10 was the Drew Brees interview at the Intel booth. The company is working with the NFL to improve helmets and reduce injuries from collisions on the field.
Intel is doing research into augmenting football helmets with sensors to determine when a player is likely to have a concussion. That way the player can be taken out of the game, and injuries reduced. The question was put to Drew, after this video was shot: “Would you take yourself out of the Superbowl if sensors showed you had a concussion?” I give you one guess at his answer. No.
My camera wigged out during the talk, so a tip of the hat goes to Vizworld for pointing us to this video.
It’s been almost a month since SC10 ended, and I’m still catching up on all the interviews I did at the show. And as I go through my notes, one theme keeps popping up: tackling complexity.
Every season, HPC users have contend more nodes, cores, and threads if they want to scale application performance. And while the programming tools are coming along, parallel programming at this level is not the forte of many scientists who just want to get their work done.
Enter Numascale, the Norwegian technology company that enables you to build an SMP out of commodity Opteron servers using their ccNuma and Numa low latency shared memory interconnect. It may sound complicated, but according to Numascale’s Einar Rustad, it’s all about programmer productivity.
Effectively, what we deliver to the end user is an SMP with scalable, cache-coherent, shared memory. With a single system image, it’s much easier for users to program, analyze, and optimize their code. In my experience, MPI programs tend to have twice the number of lines per code than are needed on a shared memory machine. You need PhD-level guys to do that kind of message-passing code, but undergrads can easily handle coding for SMP.”
Rustad went on to day that another advantage of the Numascale SMP solution is that it can run any OS for the x86 architecture, including unmodified distributions of Linux. From the OS’s point of view, its just a bigger machine. How big? With Numascale, you can build a system with up to 256 Tbytes of DRAM. And to speed applications by optimizing data locality, each Numascale card has up to 4 Gbytes of cache for storing remote data. This cache-based solution offers significant performance advantages over software-based solutions.
Call me impressed. Check out Doug Eadline’s whitepaper on Numascale technology for a closer look.
In his recent blog post reflecting SC10, John Dean of Syncsort writes that advent of parallel programming tools like Visual Studio 2010 are paving the way for HPC to reach the missing middle:
I believe this is a significant breakthrough and that the application of this technology is bound to be pervasive. It finds use in multi-core engines and in cloud applications. While it is argued that HPC and cloud are at opposite ends of the spectrum, the reality is that parallelism is at the heart of both operations. Both require function segmentation with complete definition to facilitate scalability. With the advent of cloud computing and its ”pay as you use” business paradigm, the availability of HPC clusters is becoming a reality for all of the HPC community, not just the elite few. This has the potential of bringing another 55 million users online within the next few years.
Dean goes on to say that with these developments, he can easily see HPC becoming a larger part of mainstream corporate America in the near future.
In this video from the Adva Optical Networking booth theater at SC10, Obsidian Strategics CTO David Southwell presents the challenges of interconnecting data centers over long distances using InfiniBand Transport. Recorded November 15, 2010 at SC10 in New Orleans.
I only had time to sit through one Nvidia presentation at SC10, but they must have heard me because they have now posted slides from all the talks in their booth. Highlights include:
SC10 Analyst Crossfire – Supercomputing Wrap-up Show from Rich Brueckner on Vimeo.
In this wrap-up video from SC10, Addison Snell from Intersect360 Research moderates a panel discussion on the latest developments in HPC with two industry experts and two supercomputing center directors. Recorded November 19, 2010 at SC10 in New Orleans.
Having trouble viewing? The Vimeo hosting page has a download link in the lower right corner (it varies based on your geography).
Panelists:
Topics:
Speed round:
The SC10 site has posted a video of the Clayton Christensen Keynote on Disruptive Innovation. The video is in the M4V mpeg format, so Click to download.
The TOP500 list is like chicken; there are a million potential recipes from this single source for preparing your unique dish. And so it continues today with Data infrastructure provider DataDirect Networks (DDN), who announced that the company has significantly increased its adoption in the upper ranks of the TOP500:
DDN is an established vendor in the HPC market, with a set of solutions that are tailored for the demanding needs of HPC environments,” said Earl Joseph, Ph.D., IDC program vice president for high performance computing. “Given the company’s product focus and expertise in HPC, their growth in this area is not a surprise.”
In this video from SC10, Adaptive Computing CEO Michael Jackson describes why the highest-ranked systems on the TOP500 uses the company’s MOAB system management software.
At SC10, Adaptive announced significant product upgrades that address the increasing complexity and size of next-generation HPC systems. Moab 6.0 and Moab Viewpoint 2.0 improve and simplify the command communications and reporting processes that have traditionally restricted the usability of HPC systems scaling beyond a few thousand nodes. The upgrades will undoubtedly help Adaptive Computing maintain its leadership position managing the world’s most advanced supercomputing systems as ranked by the Top500 list. Moab 6.0 and Moab Viewpoint 2.0 will be available in December 2010.
In this video from SC10, Michael McCool demonstrates Intel® Array Building Blocks, a generalized vector parallel programming solution that frees application developers from dependencies on particular low-level parallelism mechanisms or hardware architectures.
Intel® Array Building Blocks are comprised of a combination of standard C++ library interface and powerful runtime. It produces scalable, portable, and deterministic parallel implementations from a single high-level source description. It is ideal for applications that require data-intensive mathematical computations such as those found in medical imaging, digital content creation, financial analytics, energy, data mining, science and engineering. Intel® ArBB is a component of Intel® Parallel Building Blocks, and complements other Intel developer and parallel programming tools. Intel Array Building Blocks is the combination of Intel’s Ct Technology and RapidMind technology.
A free beta download is available at: http://intel.com/go/arbb
In this video, Addison Snell, CEO of Intersect360 Research, provides an overview of the HPC Market.
In part two of this video, Earl Joseph from IDC provides an overview of the HPC Market and IDC’s recent report: Strategic Agenda for European Leadership in Supercomputing — HPC 2020. Recorded Nov. 15, 2010 at the SC10 conference in New Orleans, LA.
In this video, Scalable Informatics CEO Joe Landman demonstrates the new siCluster NAS product, which is designed to scale up your capacity, storage and network performance, without scaling out of your budget.
Scalable Informatics is excited to have delivered the first siCluster NAS unit to a government customer to provide scale-out NAS storage for their computing infrastructure,” said Dr. Joseph Landman, CEO of Scalable Informatics. “The system design allows for simple management and expansion of capacity and bandwidth as the end users require. End users can connect clients to siCluster NAS using NFS, CIFS, and GlusterFS clients.”
With capacities starting at 10TB and moving rapidly to the many Petabyte region, siCluster NAS scales down and up in capacity as needed, while maintaining excellent economics. Each Delta-V storage node adds up to 32TB usable capacity, and 800MB/s bandwidth. A 96TB usable capacity system has a list price under $1,000 USD per Terabyte. This system configuration provides 6x 10GbE or QDR InfiniBand ports, and 12x 1GbE ports. Optional top of rack 10GbE or InfiniBand switches are available. Learn more at: http://scalableinformatics.com/sicluster
In this webcast, Tim Phillips and Dan Olds from the Register talk to representatives from NERSC (US National Energy Research Scientific Computing Center), Cray, and AMD in order to understand the process and learn the ins and outs of large-scale HPC.
You’ll need to register to watch the webcast, but the content is well worth a listen.
Jeffrey Layton at Linux Magazine takes an in-depth look at two major data storage trends from SC10 in the form of really fast and really dense devices.
One of the basic tenants of HPC storage is density. The more storage per rack unit (TB/U) results in fewer racks being used, shorter cable lengths, and possibly reduced costs since the number of chassis is reduced. Several companies have high density storage chassis including DDN, NexSAN, and Scalable Informatics. At SC10, LSI announced a new high density storage unit, the Engenio 2600-HD (previously known as Wembley). All of these storage devices increase the storage density by creating new chassis designs. Some of them mount the drives vertically through the top of the chassis. This allows the chassis to increase the density by putting in more drives. This also means that you have to pull out the chassis to replace a specific drive but they are usually designed for this while they continue to operate.
When it comes to speed, Layton goes on to describe the various storage tiers that HPC users employ depending on how often they need to access their data:
At SC10, the number of vendors discussing tier-0 storage technology was much greater than the last few years. This trend is significant because tier-0 storage is fairly expensive and usually he purview of vertical markets, such as the financial industry, that need the fastest performance regardless of the cost, or for large HPC centers that have very large applications and need high performance I/O to avoid creating a bottleneck and becoming a detriment to their performance.
insideHPC.com is a production of insideHPC, LLC. © 2006-2011 Sitemap