Archive for the 'HPC' Category

Hybrid Programming for Shared-Memory and Clustered SMP Systems

There’s an upcoming workshop at the PSC September 8 - 11, 2008

This workshop will present programming models and techniques for writing efficient parallel code on contemporary and future supercomputers with extensive shared memory, or hierarchical architectures with smaller shared-memory components. Two important examples of systems to which these techniques apply are the SGI Altix and networked clusters of multicore processors. Expert instructors from PSC and SGI will review MPI, OpenMP, and hardware architecture prior to launching into detailed treatments of programming for hybrid parallelism, performance analysis, and optimization. This is a “bring your own code” workshop. Participants are encouraged to bring an application to focus on during the hands-on sessions to maximize the workshop’s effectiveness. Examples will be provided for participants who cannot bring a research code. Experienced PSC computational scientists will provide support regarding the topics covered, including hybrid algorithms and implementation strategies and performance engineering.

More details

High-performance data appliances (Netezza)

This afternoon I sat through a presentation from a few guys at Netezza. They were here to discuss their system for high-performance data analytics. What they’ve effectively done is build a large database machine with some special hardware to accelerate database queries via parallel processing nodes. These are some notes I jotted down:

Architecture:

  • SMP Host
  • 100+ specialized processing units per cabinet (they named them SPU’s for “snippet processing units”)
  • SPU’s have their own PPC CPU, commodity disk, memory, and an FPGA
  • GigE networks between SPU’s
  • SMP Host partitions queries and broker activity to the processing nodes
  • Hardware fault-tolert (SPU’s can be hotswapped)

I’ll admit my skepticism tends to mount against any speaker that spends a lot of time at the outset with a marketing pitch when the audience is full of scientists. Do scientists need to be reminded that data sizes are growing? Or that enterprise X, Y, and Z are already using your product? Just show me how at works.

I did a quick search across my feeds to see if anyone has written about Netezza and (not surprisingly there is a post over at Computing at Scale. It appears there are similar efforts from Teradata, Greenplum, and DATAllegro in this space.  I can imagine how a systems like Netezza’s might complement more traditional supercomputing.  There’s certainly a big effort to commercialize the “new era of HPC” but the technologies that come out of it are business-driven and not science-driven.

What would you do with a million CPU’s?

ps3folding

There’s a new podcast on Futures in Biotech with Dr. Pande from Folding@Home. Macresearch summarized it well:

  • How a bunch of Sony PS3s have become the largest component of the world’s fastest computer
  • The challenges of distributed computing, and in particular how data storage and CPU usage can actually complement each other
  • After the hype in the 80s around computational modeling of protein structure, the computational power available today could finally make that hype a reality
  • How to take a non-parallel task and transform it into a series of computational chunks (a.k.a. how to make a baby in 1 day with 270 women)
  • How modeling of protein structure will be able to get more into the dynamics of protein conformational changes
  • What would you do if you had 250,000 CPUs?
  • I really like the final point, “What would you do with 250,000 CPU’s”, because it’s an important question. Petascale computing has arrived but most applications aren’t ready to scale to thousands or millions of cores. Folding@Home is as a distributed computing project as it is biomedical. What they’ve been able to do is treat simulations as data and use bayesian data mining techniques to put together the whole picture with suprising efficiency. A clever workaround for Folding@Home’s “supercomputer”, which is severely limited by network latencies and individual agents with slow hardware compared to ‘real’ supercomputers. Finally he reports that PS3’s and GPU’s are achieving 20-30x acceleration. Exciting stuff!

    image taken from Flikr, CC licence

    Most Innovative Use of HPC in Life Sciences

    The PSC and NRBSC have made the news again, this time in HPCwire. They’ve posted the Readers and Editors Choice awards for SC07 and the WiiMD demo earned us “Most Innovative Use of HPC in Life Sciences”.

    wiimd_bowling

    Further Reading:
    WiiMD: Bowling on Big Ben
    Engadget: wiimote used in buckyball bowling and other educational simulations

    WiiMD: Bowling on Big Ben

    SC07

    SC07 is in full swing in Reno. I promised some details about the PSC booth, more specifically the Wiimote controlled Molecular Dynamics demo. Some credit for this idea should go to Adam Marko, who is now a graduate student at UCSF joining the lab of Andrei Sali.

    The real selling point of the Nintendo Wii video game console has been the Wiimote. For those who haven’t experienced one, the Wii Remote is very different from a typical video game controller. It has a built in accelerometer, motion sensing, and optical sensing. It communicates over bluetooth, and best of all it only costs around $50 US. These features work to enrich video games through unique human-computer interaction. The sports games that come bundled with the Wii demonstrate how previously mundane gameplay such as bowling, baseball, and tennis can be a blast with clever interaction design.

    In the context of biotech, haptic interfaces could really have a big impact in how biologists, life scientists, and medical profesionals interact with computers and devices. The computer geek in me wants to believe that a keyboard and mouse is the one true way to use a computer but the success of video games consoles, Tablet PC’s, and iPhones are changing the face of computing. Imagine biosimulations which are no longer just scripts, plots, and post-process rendering but rather augmented reality. Where a scientist can use virtual instruments and interact with their simulations in real time.

    Before I get ahead of myself, let me explain the demonstration you can see at the PSC booth if you’re attending SC07. It is quite simple to use a Wiimote to replace your mouse on a standard PC. Several months ago we played around with Wiimotes on our workstations, manipulating protein structures in VMD and CHIMERA which are both excellent molecular graphics and modeling programs. In this case the Wiimote just acts as a mouse, and doesn’t really offer much advantage, especially for someone who uses VMD with a mouse and keyboard. Pretty cool, but not very practical for me personally. So the Wii Remotes were left in the drawer until discussion about demos for SC07 arose. That’s when some other PSC heads got involved and the idea blossomed.

    VMD is a cool program to use with a Wiimote, but a moving a single protein structure isn’t very interesting. Let’s do a NAMD simulation instead. Better yet, let’s use IMD to have an interactive NAMD simulation. And while we’re at it, let’s run it live on Big Ben (PSC’s 4,000+ processor Cray XT3). After all, Supercomputing is our middlename. The “WiiMD” demo is live at SC07 thanks to key efforts by Nathan Stone, Shawn Brown, and student Jordan Soyke. Unfortunately I could not be there to see the final result, but some work was done with PDIO, a middleware that routes data straight from the nodes of the XT3 to a remote location in real-time. There was also a rewrite of the Wiimote input driver, which Jordan talks about briefly in his blog.

    Stop by the PSC booth if you get a chance. I hope to have pictures or video of the demo up sometime soon.

    PSC Live! at SC07
    WiiMD, interactive molecular dynamics with a wiimote?

    And my two favorite blogs have noticed!
    scalability.org - Coolest demo I saw today
    BBGM - Finally found a reason to buy a Nintendo Wii

    Supercomputing 2007

    It’s that time of year. SC07 begins in Reno next week, and everyone around here is getting prepared. Unfortunately this is the second year in a row that I can’t attend Supercomputing. Last year I was at CASP and this year we are doing a 2-day structural genomics workshop at UTSA. Apparently I’m not the only one with Supercomputing blues. I will miss the chance to talk with vendors and work the PSC booth, but I at least have a poster and some labmates going to Reno.

    While much of the conference is focused on high performance network technologies, there’s some great life science talks worth attending. There’s also a session that grabbed my attention,

    High Performance Computing on GPUs with CUDA

    where NAMD and NVidia developers will demonstrate MD simulations on graphics processors. Stay tuned for some juicy details about a demo from our booth.