HP Scales Out With New Cloud Servers
HP is stripping down to scale out. Today the company is launching a line of servers and services targeting the cloud computing market, which is accounting for a growing chunk of server sales. HP’s Extreme Scale-Out (ExSO) portfolio is designed to appeal to cloud-builders’ focus on energy efficiency and cost, and be delivered by the rackload.
HP’s ambitions are built around the ProLiant SL server line, which features a “skinless” architecture with a lightweight rail and tray design. The new servers are deployed in HP’s new ProLiant z6000 chassis, a 2U frame housing shared power supply and fans. The common chassis allows customers to choose between three ProLiant SL configurations optimized for maximum memory, storage or processing power.
On each server, HP is mounting the disks on the side rather than the front, allowing better air flow through to the fans on the rear of the chassis. The i/o cabling is on the front of the server, allowing it to be serviced from the cold aisle.
“We’ve designed it for power efficiency and good airflow,” said Ed Turkel, manager of business development for HP Scalable Computing & Infrastructure. “In this environment they typically don’t need a lot of management features. We think of this as a lean and mean environment.”
“Customers with scale out business models need solutions that make every dollar, watt and square foot in the data center count,” said Christine Reischl, senior vice president and general manager of Industry Standard Servers, HP. “These customers are pioneering ‘extreme’ businesses and their extraordinary requirements are pushing beyond traditional architectures.”
HP is hardly alone chasing the cloud computing market, and enters a niche where rivals IBM (iDataPlex), Dell (CloudServer) and Rackable (MicroSlice/CloudRack) have already introduced servers tailored for customers with scale-out data centers.
HP is touting the ProLiant SL’s ability to put data centers on a diet, both in terms of energy consumption and the actual weight of the racks. The SL line with Intel Xeon 5520 processors uses 28 percent less power than the HP ProLiant DL160 Server, and its skinless design makes it lighter. HP says the new SL2x170z is about half the weight of traditional rackmount servers, offering savings on shipping costs and allowing lower floor support requirements.
Here’s a comparison of the models and features:
- HP ProLiant SL160z: This server is optimized for workloads with large memory requirements, with 18 DDR3 DIMM sockets (compared to 16 for the other SL models). The SL160z also has two PCI slots to support large memory-cache apps.
- HP ProLiant SL170z: Storage capacity is the focus of this configuration, which offers 6 large-form Serial ATA or serial-attached SCSI drives to support storage-intensive workloads like web search and database applciations.
- HP ProLiant SL2x170z: With two servers in each 1U tray, this model is designed with extra compute power for high-density HPC environments and web front-end applications.
HP says the ExSO portfolio is ideal as the inner workings of the HP POD data center container. ”We’re talking to some customers about deploying by the POD,” said Turkel. “There’s a lot of interest in that. But we expect most customers will want to deploy these servers in a traditional data center.”
HP is also introducing ”Extreme Operational Support” for ExSO customers, which it says provides installation services specifically designed for data centers with thousands of servers. ExSO customers can also opt for an on-site “seed inventory” of replacement parts to speed repairs.
While much of the scale-out focus is on cloud computing, HP’s new product line was welcomed by the company’s partners targeting high-performance computing (HPC), including Microsoft.
“HP’s new Extreme Scale-out portfolio directly addresses a number of key challenges faced in today’s mega-data centers, such as efficiency in power and cooling, while maximizing performance and density per square foot,” said Vince Mendillo, director of HPC marketing for Microsoft.
This is interesting. I though it was all about blade servers these days. I guess the cloud has put new life into rack mount servers.
cisco_guruPosted June 10th, 2009
even Cisco is annoucing rack-mounted server, the C-series. This to allow more entry-level customer to benefit of Unified Fabric, Extended Memory and so on that UCS blade offers today.
cmholmPosted June 11th, 2009
Looks like someone found religion after Google released photos of their server-on-a-sheet form factor.
I wonder if they will strip the casing off their blade servers. This kind of idea seems like a no brainer. But are they reducing the insane amount of packaging that a single fully populated blade center sheds on installation? Its basically a full skip bin worth of wood, cardboard and plastic for a 1/3 rack of servers.
Cisco UCS = over priced enterprise servers. It would not be caught dead in a scale out datacenter.
HP finally figured out that blades are no more power efficient than an optimized standard rackmount server, something their scale out customers have been telling them for over five years. While the SL6000 is getting all the press the same half width system boards are used in the DL1000 which will likely have broader appeal. The SL6000 is targeted at low margin high volume customers and are only a good fit for Microsoft, Yahoo, AIC, Facebook, and the like.
oldskoolPosted June 12th, 2009
Yeah but the cooling is still front-to-back. VME and Compact PCI mount cards vertically so you can flow the air right from your under the floor aircon all the way up through the stack.
DrewPosted June 12th, 2009
Blade servers are pretty useless for large deployments (e.g. deployments that fill a whole building). Most datacenter shells are built for 200-250 watts per square foot maximum and a lot of these designs are capable of >1000 watts per square foot. So I don’t really know who they’re aiming for with these super high densities, but not the Yahoos and Facebooks who know the density of their server rooms.
These servers really aren’t that interesting – they are basically just slightly larger blades. Proprietary motherboards means you’re locked into HP for the form factor. The Rackable CloudRack is kind of cool because you can put different ATX motherboards in there, but it’s still too much case. What would be a huge win is if someone designed a rack with a common AC/DC converter and an ATX power cable for each of 100 motherboards, and had some drive slots and big fans in the whole thing. No extra metal, super cheap. Even a cheap rack plus 2 PDUs costs around $2000, plus 40 or so chassis and power supplies cost $8000+.. so anything you could charge under $10k for the rack and power distribution itself would be a cost saving.
rajeevPosted June 12th, 2009
I was waiting to hear somethign like this from HP ..only sad point is ..they have copied 100% IBM iDataPlex so it looks like a mirror image of a exsisting solution . I was expecting something more innovative..something better then iDataPlex and Cisco USC solution.
yeh i’m still waiting for someone to explain to me why distributed power isn’t more widely deployed.
i have 5 servers at home and it just seems like a given to power them from a single power supply?
even better if it incorporate the functionality that my UPS supporting these servers does.
DefenestratorPosted June 14th, 2009
Part of the problem is that most datacenters aren’t run by the people actually using them. Other than a few big players, most datacenters are set up so that each customer can drop in a rack or partial rack of 1U/2U/whatever servers that will be plugged into CAT5 and 120V AC. It’s hard for something like per-rack power conversion and UPS to take hold when most datacenters already have a large UPS already built and running.