On Wednesday HP announced a new focus on what it is calling extreme scale with its Extreme Scale-Out (ExSO) portfolio. The new offering is a collection of hardware and services targeted at those, whether in very large Internet application businesses or in high performance computing, who crave quantity.

And by “quantity” HP means those building clusters of thousands to tens of thousands of servers. HPCwire dug into the announcement in a discussion with HP’s Ed Turkel, who explained that this launch represents “a new focus in extreme scale-out computing driven by the needs of the Web and high performance computing communities.”

For starters HP has introduced the ProLiant SL server family with a “skinless” design that basically puts the server component on a sheet metal tray (not unlike the Google design) rather than enclosing it in a metal box. They further reduced the metal used in the design by cutting out sections of the tray, further reducing weight, shipping costs, and environmental impact. In fact, the SL series is about 30 percent lighter than “standard” rack servers. The SL puts up to four dual-socket Nehalem servers in a 2U unit (672 cores in a 42U rack), and adds high-efficiency fans and power supplies that are all shared by the servers on a tray. The airflow is front-to-back, and HP emphasizes that this design uses standard 19″, 42U racks — no need to shift things about to accommodate unusually wide or deep racks.

They have adjusted the packaging, but the motherboards themselves still support the range of components and interfaces you’ve come to expect from HP’s other high-end server lines, and the systems support 1 and 10 GbE as well as InfiniBand, which will be important to certain kinds of HPC users, especially those with the tightly-coupled applications typical of engineering and scientific applications. What is missing from the SL family is many of the RAS and management features you’d find in an enterprise-class server. For example, the SL does support IMPI, but not HP’s full-blown Intelligent Lights Out features.

HP is also talking abut the energy profile of the new system, something that anyone deploying thousands of servers is going to have to worry about. HP’s data indicate that the consolidation of power supplies and fans in each rack is saving 28 percent over traditional rack-based servers. Their analysis (details in the press release if you want to read them yourself) is based on 88,000 servers in a 100,000 square foot datacenter, and shows that in this scenario the datacenter would save as much as $4M a year in energy costs using the ProLiant SL solution.

HP emphasizes the use of standard components for its ExSO portfolio, contrasting it with solutions like IBM’s iDataPlex or SGI’s CloudRack, which use non-standard components like half-depth motherboards. From HP’s perspective the standards approach lets it get economies of scale with its other hardware lines and also gives customers the flexibility to include standard 19″ components in the racks of an ExSO system without any finagling. Despite the use of standard components, HP has not yet sought Cluster Ready (ICR) certification through Intel’s industry program. When I asked about ICR, Turkel pointed out that HP’s Unified Cluster Portfolio program essentially has the same goals, and that in spirit the two efforts are similar enough that the new offering should already be most of the way there should HP ever decide to pursue Intel Cluster Ready certification. He says that when he talks to customers what they want most is to know that the parts they are buying will work together, and customers get that from the Unified Cluster Portfolio program.

HP is also talking about the rest of your big compute solution by introducing a new engagement and delivery process on the professional services side specifically geared to large deployments. This model includes the pieces that larger customers are likely going to demand as standard components of the service, including onsite parts storage and the opportunity to do routine, scheduled visits to deal with server issues several at a time, rather than as they occur. This model recognizes that with thousands of servers customers are going to lose servers from the compute pool fairly regularly, and that because of the large scale of the system, losing a couple servers likely isn’t something that has to be dealt with as quickly as when a customer loses a server from a 16-processor system.

One of the pieces of the new ExSO portfolio that I was very interested in was HP’s new Environmental Edge solution. Environmental Edge is a hardware and software solution for monitoring the datacenter environment and the servers themselves. The product uses wireless sensors to monitor things like temperature, humidity, air pressure, and power utilization throughout the datacenter. The wireless aspect may pose a security concern for some datacenters with security teams suspicious of all things wireless, but for those with an existing (and busy) datacenter, this option is a lot more attractive than going through the disruption of a fixed-wire sensor network deployment.

Environmental Edge includes a management software component that allows center managers to take in all of the data and analyze what’s going on. As systems grow in scale and complexity, they draw an ever-increasing portion of a datacenter’s power capacity. As centers near that capacity they need to have an accurate, detailed view of how much of their power and cooling budget is consumed, and where they have excess capacity. Having this information today means knowing which circuits have available capacity on them when managers need to deploy new systems, and knowing where hotspots are forming so that cooling can be adjusted to compensate.

But once centers are more commonly instrumented and interfaces to those data are more widely accepted, HP sees a future in which environmental data are taken into account in routine operations. For example, if a job scheduler was aware of the power utilization of all the machines in a center, it could move jobs to machines with lower energy use, or even delay running those jobs until off-peak power rates made running a particularly power-intensive job cheaper. HP isn’t ready to talk much about this, other than to say that it is working on it now. It will certainly allow us to take a much more integrated approach to sustainable computing practices while still getting the scale we need to do our jobs.