Features

When blade servers burst on the computing scene several years ago, they were hailed as a replacement for traditional rack-mount solutions and a catalyst for the continuing shift away from proprietary Unix servers and mainframes and towards systems leveraging the x86 architecture and Linux. Fueled by promises of lower cost, higher density, and easier manageability, 1,000-node blade deployments were expected to address enterprise server hardware consolidation needs.

Like many new highly anticipated enterprise technologies, the idea of blade-based clusters has so far been more compelling than its reality. Blades have only taken limited market share from standard rack-mount server designs, and many early blade adopters are now abandoning blade solutions. Closed architectures, limited choices in components and software, thermal management issues, and vendor lock-in are some of the most serious shortcomings that have kept large-scale data centers from embracing blades as a replacement for rack-mount solutions.

A Giant Step Backward
The hype surrounding the blade market created a marketer's paradise: large x86 server vendors repackaged their server technologies in thin form factors, positioning these blades as a denser alternative to rack-mount solutions that were cheaper than minicomputers or "big iron." Vendors incorrectly assumed that these repackaged solutions would scale just like their traditional rack-mount counterparts.

With a shift from selling product specifications to business benefits, blade makers quietly flooded the market with machines designed to lock customers into a single supplier - a stealthy regression back to the era of high-margin proprietary computing. In the transition to this new technology, the largest server vendors failed to implement the systemic changes necessary to turn the idea of a single blade node into an overall blade-based data center solution that meets enterprise expectations of performance and compatibility.

Customers quickly recognized three primary limitations of blades:

• Proprietary hardware architecture. Traditional blade vendors dramatically limit flexibility by locking users into specific motherboards, IP networking, and Storage Area Networking (SAN) interconnects - forcing data centers to compromise choice while increasing costs.
• Improper engineering for larger deployments. Most blade solutions are engineered in relatively small chassis systems (e.g., 7U high) and contain an embedded power subsystem, embedded networking, and embedded storage interconnects (e.g., a SAN fabric). Generally speaking, the architectural choices made for such small systems are not optimal for large deployments of hundreds or thousands of nodes with pre-existing data center software automation strategies that require flexibility.
• Excessive focus on density at the blade level. This leading to poor thermal engineering, limited Direct Attached Storage options, and limited PCI expansion options. Due to poor thermal engineering, manufacturers must often use non-standard CPUs and limit processor speed options or make use of the thermal throttling features inherent to some CPUs. The limitations in drive selection and PCI slot availability impact the flexibility of building configurations with fast local storage or expansion cards.

Although the overall blade market continues to grow for small-scale deployments, the limitations found in blade server designs will continue to keep large data centers from considering solutions that simply can't scale to meet their growing needs.

Real Needs Demand Real Solutions
The need for scalable data center server solutions is stronger than ever as pressure mounts for enterprise IT departments and data center managers to cut costs and complexity. As monolithic proprietary Unix servers and mainframes continue to lose market share to x86 Linux-based solutions, enterprises desperately search for solutions that will:

• Enable higher server density and better use of data center floor space
• Adopt open standard components at the motherboard, CPU, drive, and PCI expansion level
• Simplify deployment and serviceability
• Be properly engineered for very large deployments, including the rack/cabinet or data center level
• Offer a variety of power distribution strategies
• Offer a choice of IP networking and SAN interconnect strategies, including access to and compatibility with full product sets from leading vendors such as Cisco and Brocade
• Work with a variety of software fabrics from leading vendors, enabling basic data center automation strategies, server consolidation, various kinds of clustering, and high-availability levels

The demand for these specifications echo from a wide range of high-performance and enterprise computing environments in industries including oil and gas, biopharmaceuticals, entertainment and rendering, government labs, large-scale Web farms, and very large Fortune 500 and financial services deployments.

The industry needs solutions that can deliver the core benefits of traditional blades - serviceability, ease of management, and density - without compromising access to industry-standard components. Additionally, these alternatives should feature the flexibility and innovation necessary to reduce or largely eliminate thermal management issues while enabling total choice in hardware interconnect, operating systems, clustering layers, and management software. The newest generation of rack-mount server designs meets these specifications.

The Building Blocks of Large-Scale Clusters
Today's rack-mount server solutions feature optimized form factors, superior cooling innovations, and rack-level deployment strategies designed to leverage the best aspects of traditional blades without the obvious challenges of thermal management, scalability, and vendor lock-in. Designed from a rack-level, data center perspective to meet the real-life needs of customers wanting to deploy large-scale turnkey solutions quickly, the latest breed of rack-mount solutions provides a solution superior to the much-hyped blade.

With the introduction of the half-depth form factor in 1999, rack-mount servers could be mounted back-to-back inside a cabinet for twice the density of full-depth servers. Back-to-back mounting enables data centers to increase density and improve airflow radically, with air entering the front of the servers and exhausting to a central plenum inside the cabinet. Serviceability was also improved by the introduction of more efficient cable management and front-facing I/O. Administrators can more easily service and manage individual nodes with the same characteristic of blade designs.

Building a Better Blade
In 2004, the industry's first hybrid rack-mount/blade solution was unveiled. This new form factor delivers the core benefits of traditional blades - easy serviceability, management, and density - in conjunction with the best features of rack-mount server designs. Dubbed the "open blade," benefits include the use of industry-standard components, including today's fastest processors from AMD and Intel at the highest thermal envelope, along with superior power management and heat evacuation.

The form factor rivals and even exceeds the density levels of traditional blades, featuring back-to-back and side-to-side mounting of up to 92 nodes per cabinet with an additional 10Us available for networking gear. Cable management is simplified with cables that are internally routed to a backplane-mating scheme so no server-level cables are exposed. Tool-less design enables the easy insertion and removal of servers.

Equally important as density and serviceability is the issue of thermal management. With blade users reporting heat output as high as 14 kilowatts per rack (equivalent to the amount of heat required to cook a small turkey!), the issues of overheated components and reduced server reliability are very real. Back-to-back mounting achieved with open blade rack-mount servers lets heat be evacuated more effectively through a central air plenum, helping to reduce the hot spots common to traditional blades.

Optimized rack-mount designs also take full advantage of longer heat sink fins - which are typically much smaller in blades due to their skinny form factor - to better cool processors and power supplies. Up to 92% of the airflow in open blade servers is directed over the processors for optimal cooling, resulting in longer server life expectancy and reliability.

Finally, these open blade rack-mount designs allow for full flexibility in hardware interconnects, as well as operating systems, clustering layers, and management software. They support an open architecture approach and the industry trend toward open source Linux-based cluster solutions.

DC Power: Greater Reliability
The industry's most progressive rack-mount solutions have been carefully designed to manage heat and reduce thermal issues more effectively. Yet the introduction of DC power-based servers in 2003 has had one of the most significant impacts ever seen on server reliability and overall data center cost savings.

Unlike traditional blade servers, today's leading rack-mount and open blade servers can be powered with a vastly more reliable DC power supply. DC power supplies achieve efficiency levels of 93% at the node level, compared to 70% efficiency in traditional AC power supplies. By distributing redundant DC power to each server, approximately 20% to 40% of the thermal load is shifted outside the server to AC-to-DC rectifiers. As a result, server reliability is increased by as much as 27%, and monthly power costs drop by as much as 30%.

Because today's most advanced DC-based servers are compatible with both AC- and DC-based data centers, the switch to DC servers is an inevitable industry-wide trend as demand for larger clusters puts new burdens on power requirements, HVAC limitations, and IT budgets.

The Operational Advantages
As today's newest rack-mount designs address the demands of organizations that need hundreds, if not thousands, of data center servers, they help enterprise data centers regain much of the operational efficiencies lost with traditional proprietary blade clusters.

By improving the flexibility of deployed infrastructure with simple plug-and-play servers, the node becomes a field-replaceable unit in today's increasingly cluster-centric environment for the first time. Such simplified installation can deliver significant savings of both time and money, while addressing the concerns of combining available space, power, and cooling capabilities in a large data center. Today's novel rack-mount solutions are flexible enough to make good on the operational advantages once promised by blade-centric server vendors.

Positive Financial Impact
One of the most significant advantages in leveraging rack-mount solutions is that they can dramatically improve a data center's financial dynamics. A pay-as-you-grow model of server expansion offers incremental capital outlays instead of large-scale forklift upgrades. Enterprise IT departments get exactly the computing power they need, when they need it.

The ability to scale this way is propelled by the continued growth of Linux and the scalable software stacks designed to run on it. The proliferation of enterprise-grade operating systems running on commodity x86 architectures has encouraged the development of software layers that take advantage of multiple one-, two-, and four-processor servers. The growth of Linux-based solutions in the data center will continue to drive this trend to scale-out computing.

Because rack-mount, open blade designs are based on an open architecture approach leveraging a wide array of industry-standard components, enterprises can achieve additional savings by selecting vendors of their choice - avoiding the premiums associated with proprietary equipment and vendor lock-in. These features speed time-to-market or time-to-service delivery while enhancing reliability and reducing the costs associated with cluster configuration and management. From installation to ongoing management in Linux-based clusters large and small, these solutions can directly and significantly lower costs while providing enterprise-class computing power.

Are Blades Here to Stay?
Despite the limitations of traditional closed blades, their sales soared more than 650% in 2003, driven largely by massive marketing campaigns. Gartner, the analyst firm, projects that over the next five years blade servers will account for nearly 16% of worldwide server shipments, reaching a compound annual growth rate (CAGR) of 30.6% by 2009.

But what is truly in store for the blade market? Will it meet and beat analyst expectations? Or is it going to run its course and let rack-mount servers take back their original territory? Featuring previously unheard of technological and financial benefits, today's novel rack-mount server designs may well be the next big step, supplanting the dramatic growth of closed blade servers. Or will another kind of blade hybrid burst on the scene? It remains to be seen, but for now companies are working to uncover the ideal solution that will help them build a perfect data center. Facing unprecedented demands, rack-mount open blade hybrid designs may be just the right fit for the times.

Add Your Feedback

In order to post a comment you need to be registered and logged in.

Register | Log in

Please wait while we process your request...