Features:

RACKABLE CEO DISCUSSES NEW 'OPEN BLADE' SERVER SERIES
by Tim Curns, Editor

Rackable Systems recently unveiled their ‘open blade’ alternative server series, built with a rack-level, data center perspective for large, scale out data center deployments. This ‘open blade’ alternative server incorporates back-to-back rack mount server designs with the best features of blade servers. In return, this open architecture approach provides the industry’s highest server density, maximum compute power and most flexible configurations ­ without compromising ease of serviceability, maintenance and management.

The solution is designed for the large segment of the server market that demands high density and high server counts. The Scale Out Server Series is built from a rack-level, data center perspective as opposed to the 7U or sub- 7U chassis design typical of traditional blade server solutions. Traditional blade design inhibits open architecture design while severely hampering thermal efficiency. Scale Out's open architecture approach maximizes non- proprietary, industry-standard components, emphasizing unlimited flexibility for customers requiring specific networking, local storage, storage interconnect and provisioning software solutions.

HPCwire had a chance to speak with Rackable's CEO, Tom Barton, about this development.

HPCwire: How will the open-architecture empower large-scale, high density data center deployments?

Tom Barton: Open architecture solutions empower large-scale, high density deployments in many ways:

• enable access to a larger range of components: full range of ATX and Extended ATX motherboards; leading CPUs, various DRAM types and vendors; best of breed server-grade hard drives, etc.
• open standard components tend to be lower cost vs. lower-volume proprietary components
• enables greater architectural flexibility and greater access to best of breed IP networking and storage interconnect (NAS, SAN) solutions.

Leveraging open architecture and industry-standard components, Rackable Systems' Scale Out solution has been designed to provide very high density with 92 compute nodes that feature leading-edge thermal performance. Each node can support up to 2 high wattage CPUs-for example, two 120W 3.8GHz Intel Xeon EM64T (Nocona) processors. Even at this high density level with maximum compute power, per-node options are not sacrificed and can include up to two server-grade hard drives and a full-height PCI expansion card via a right- angle riser.

HPCwire: What benefits come along with the "open blade" alternative? How will this affect the market in general?

TB: The 'open blade' design of the Scale Out Server Series combines the best features of traditional blades with the best features of Rackable Systems' rack-mount servers. Key benefits drawn from blades include plug-and-play modular computing; integrated mating of the blade to the cabinet; internalized cable management; and ease of serviceability with no tools/screwdrivers required.

Yet traditional blade solutions tend to be thermally-challenged, have lower density than the Scale Out Series and are hampered by proprietary architectures that restrict options around power distribution, PCI expansion, motherboard feature sets, CPU choices, networking choices and storage interconnect strategies. The Scale Out Series eliminates all these issues by providing total component and best-in-breed software choice from a broad range of vendors, along with leading thermal management and cooling technologies that support up to 184 processors at the highest thermal envelope from any vendor's roadmap.

The Scale Out Server Series provides a new option for customers seeking to 'scale out' their data centers with highly dense clusters, without being locked into proprietary blade solutions.

HPCwire: You just won a LinuxWorld award for Most Innovative Hardware Solution. Why is your announcement so important? How does it differ from typical blade structures?

TB: The Scale Out Series marks an evolution in data center server technology, and represents a very well engineered, sensible blade-like structure for large, 'scale out' deployments that provides access to lower cost, open standard components. In addition to the key differentiators mentioned above, one major difference is that most competitor blade structures are inherently optimized around smaller deployments, e.g., a 7U chassis with 14 blades. The Rackable Systems Scale Out series has been engineered at the data center/rack level-for those deploying 92 systems at a time.

HPCwire: How will the data center be cooled with your structure?

TB: Like all Rackable Systems servers, the Scale Out line has been engineered to provide the most thermally-efficient solution possible. With servers mounted back-to-back and side-to-side, heat is encapsulated in a central plenum inside the cabinet, and then exhausted out the top. For data centers with air returns in the ceiling, the heat is vectored very efficiently to the air return, minimizing general heat dissipation in the data center, thus keeping the aisles cooler. For those customers choosing data center level DC solutions (e.g., telco DC power distribution units), up to 40% of the heat can be removed from the data center by spot chilling the PDUs or locating the AC/DC conversion outside the server HVAC envelope.

HPCwire: Will this new architecture compromise ease of serviceability, maintenance and/or management?

TB: No. In fact, it offers even better serviceability, maintenance and management than competing blade structures. For example, the IBM BladeCenter is built as a 7U chassis, meaning that multiple 7U chassis have to be mounted in a cabinet or in telco 2-post racks to achieve required density levels. By contrast, our Scale Out servers are deployed as a rack-level solution-fully populated, racked, cabled and ready to be wheeled into the data center and connected to power and network uplinks. Rackable Systems deploys at a cabinet level vs. a chassis level.

The Scale Out Series has completely internalized cable management with no externally exposed cables. Systems slide in and out, and networking, remote management and power connectivity is automatic, similar to competing blade structures.

Regarding management, the Scale Out Series is available with Rackable Systems' innovative, lights-out serial remote management solution, and can optionally be enabled with IPMI based remote management. Rackable Systems offers open APIs and a scripting mode that enables easy integration with best-of-breed data center automation fabrics such as Altiris, BladeLogic, Veritas OpForce, Oracle 10g, Platform Computing Platform LSF, etc.

HPCwire: What else is Rackable working on to propel its position in the blade service industry?

TB: Rackable Systems is examining a range of technologies that may be relevant to advanced backplane and interconnect strategies, such as 10GigE and Infiniband.

HPCwire: Thanks for sharing your insights, Tom. And congratulations on earning the award for Most Innovative Hardware Solution at LinuxWorld!