HPC Features in New ProLiant Gen8 Servers

This week HP announced the new ProLiant Gen8 server line, the world’s most self-sufficient servers. I know, HPC customers are looking for the detailed speeds and feeds, and that information will become available in the coming weeks as specific ProLiant Gen8 servers are announced. However, there is plenty of information today to talk about with customers looking to deploy a complete HPC solution. Today I’ll review four key pillars of the ProLiant Gen8 architecture:

  • Integrated Lifecycle Automation
  • Dynamic Workload Acceleration
  • Automated Energy Optimization and
  • Proactive Service and Support
    and explain how HPC customers benefit from the millions of dollars of research and development that went into HP’s newest servers.

    Integrated Lifecycle Automation

    Managing a cluster of 100’s or 1000’s of compute nodes can be a challenge for any customer. At the heart of HP’s industry leading management capabilities is the HP iLO (integrated lights out) management engine. With over 28 million ProLiant servers deployed in the market, iLO is in such widespread use that HP’s iLO drivers have been incorporated into the baseline kernel.org Linux code base and thus included in all Linux distributions. How many other vendors can claim that about their systems management capailities?

    The newest iLO4 engine included in Gen8 servers introduces several advances of significant interest to HPC customers. Perhaps the biggest is that iLO4 doesn’t require any host OS daemons to operate, thus significantly reducing OS jitter, a common cause of unpredictable application performance. Most other management tools on the market require a management process, or daemon, to run on the host OS. Such daemons typically use < 1% of the host CPU and thus have little impact on enterprise workloads. However, in an HPC environment, when you may have 1000's of nodes operating in lockstep, for instance running a cluster-wide MPI job, each time a management daemon wakes up and runs it not only slows down that individual node by 1%, but slows down all nodes in the cluster by 1%. It doesn't take too long to wreck havoc on cluster performance. Historically many HPC sys admins have spent a lot of time removing all management daemons for exactly that reason. Of course some people drive while talking on a cell phone too. Flying blind is not a good idea.

    Now with iLO4, no management daemons are required on the host OS, simplifying installation and significantly reducing OS jitter and boosting performance. In addition, since no host OS software is required for iLO4, that means management capabilities are available as soon as you cable your server with power and networking, before you load an OS, and even if you should have an OS failure during operation. In addition, the ProLiant Gen8 Active Health Monitoring system captures critical server diagnostics completely within the iLO4 engine, without host CPU or host OS intervention, preserving the data for HP service personnel should there be any host OS or hardware failure.

    Dynamic Workload Acceleration

    Every HPC customer I’ve ever talked to is interested in higher performance, and ProLiant Gen8 features a host of dynamic workload acceleration technologies to deliver higher performance in an industry standard server. In particular, HPC applications are particular heavy users of memory and i/o subsystems. Starting with the memory subsystem, HP has engineered the ProLiant Gen8 line to support the fastest currently available 1600 MHz memory on two DIMMs per memory channel. When you add a second memory DIMM to a memory channel on many other servers, the memory system will clock down to 1333 MHz, even when using 1600 MHz DIMMs. Going beyond the memory to the storage system, ProLiant Gen8 servers introduce a new SmartArray controller. The latest SmartArray controller not only supports double the number of drives but also is optimized for new SSD/flash technology which more and more HPC customers are incorporating into their servers. The Gen8 family will also continue its industry leading support for Nvidia GPUs, expanding on the integrated GPU support offered in the current SL390s Gen7.

    Automated Energy Optimization
    The ProLiant Gen8 family starts by offering some of the most energy efficient power supplies available, with up to 94% efficiency. Many supposedly “low cost” servers have power supplies with 84% or less efficiency, leading to 10% or higher power costs which can really add up over the lifetime of your system. HP Intelligent PDUs automatically track power usage and document configurations to increase system uptime and reduce the risk of outages. And with the new ProLiant Gen8 FlexNet networking, energy management is even extended to your network interfaces.

    Proactive Service and Support

    A showcase of the new ProLiant Gen8 proactive service and support capabilities is HP Insight Online, the industry’s first cloud-based IT management and support portal. But even customers who can’t connect their HPC clusters to the outside world can benefit from service and support features designed deep into the new ProLiant Gen8 line. My favorite is the new smart disk drive carrier, not only is the carrier more compact, allowing more drives to be fit into the same size server, but a new “safe to remove” light on the carrier helps prevent you from removing a drive that is actively in use by the system.

    The ProLiant Gen8 family will include many models spanning HP’s DL (rack mount), BL (blade), and SL (scale-out) product families, but with its Integrated Lifecycle Automation, Dynamic Workload Acceleration, Automated Energy Optimization, and Proactive Service and Support, these servers are more optimized for HPC than any other industry standard server on the planet. Starting with the Purdue Carter cluster, which debuted at #54 on the Top500 list last November, HP has already deployed 1000’s of ProLiant Gen8 servers to HPC customers through our early ship program, and with this week’s ProLiant Gen8 announcement and upcoming server launches, the power of ProLiant Gen8 will soon be available to all HP customers across the planet.

    Advertisements
  • About Marc Hamilton

    Marc Hamilton – Vice President, Solutions Architecture and Engineering, NVIDIA. At NVIDIA, the Visual Computing Company, Marc leads the worldwide Solutions Architecture and Engineering team, responsible for working with NVIDIA’s customers and partners to deliver the world’s best end to end solutions for professional visualization and design, high performance computing, and big data analytics. Prior to NVIDIA, Marc worked in the Hyperscale Business Unit within HP’s Enterprise Group where he led the HPC team for the Americas region. Marc spent 16 years at Sun Microsystems in HPC and other sales and marketing executive management roles. Marc also worked at TRW developing HPC applications for the US aerospace and defense industry. He has published a number of technical articles and is the author of the book, “Software Development, Building Reliable Systems”. Marc holds a BS degree in Math and Computer Science from UCLA, an MS degree in Electrical Engineering from USC, and is a graduate of the UCLA Executive Management program.
    This entry was posted in HPC. Bookmark the permalink.

    5 Responses to HPC Features in New ProLiant Gen8 Servers

    1. Gary Edlund says:

      Wondering if you’ve tested Netlist’s HyperCloud 4, 8 16 or 32GB 2vRank DDR3 RDIMM’s on the ProLiant Gen8 Servers?

    2. Gary,
      I can’t make any comments as to what memory is supported in specific Gen8 servers since the servers themselves have not been announced. The general comments about memory latency that where made in my earlier blog post where all done with HP memory. The Netlist HyperCloud memory provides the biggest advantage for servers that don’t support full memory speeds when you configure your server with multiple DIMMs per channel. In many HP servers, we already support running all memory channels at the maximum speed using standard HP memory, and thus in those servers there is no performance gain by using Netlist’s HyperCloud RDIMMs. HP procures memory, disk drives, and other components from many vendors and selects specific components that are best suited to meet the demands of our users. As we announce specific models in the Gen8 family, especially our HPC oriented SL family, I’ll be sure to post some followup blogs on specific memory optimization options for those machines.

    3. Jo Cohen says:

      Hi Mark,
      Just curious. So you are saying that many HP servers support 16GB and 32GB running 3DPC at 1333MT/s – and those “regular” RDIMMs will run at full bandwidth on Gen 8 servers?
      That’s not the information I got from HP site:
      http://h18004.www1.hp.com/products/servers/options/memory-description.html
      I am also aware of only 1 2-rank 16GB from HP, and it also does not seem to run 3DPC at 1333MT/s.
      In addition to its 16GB HyperCloud, NLST claims that its 32GB HyperCloud will run 3DPC at 1333MT/s, something that Samsung and MU 32GB LRDIMM cannot do. Also few of the LRDIMM sites I encountered confirm that fact (SuperMicro, Samsung brochures).
      Otherwise, generally speaking, I thought that memory subsystems require Load-Reduction to be able to sustain 3DPC at full speed.
      Thanks

    4. Jo Cohen says:

      Mark,
      I think you may not aware that HP and Netlist have signed an “exclusive” collaboration agreement 3 months ago with respect to qualification and marketing of Nelist HyperCloud memory.
      Also, currently there’s no known way that regular 32GB RDIMM can even run at 3DPC – as compared to Netlist claims that its 32GB HyperCloud run at 3PDC and full bandwidth of 1333MT/s.

      Netlist also stated that its HyperCloud Load-Reduce RDIMM runs faster than LRDIMM. Any comments on that?
      Thank you.

    5. Jo Cohen says:

      BTW, here’s a Cirrascale page from 3 months ago, showing their rack mount server RM2325 supporting 288GB at full 1333MHz using HyperCloud from Netlist.

      http://www.cirrascale.com/documents/RM2325_Rackmount_Server.pdf

      I don’t think I ever seen a memory module achieving that capability, including HP own memory – at least according to HP’s own web site.

      I also saw an announcement from GIGABYTE of their GA-7TESM server capable of running 288GB at 1333MHz using HyperCloud (Sep 13, 2011)

    Comments are closed.