This is part one of our three-part series on blade servers. Here we examine the cost-saving possibilities of blades.
For hardware vendors, single-digit revenue growth in the server market has become the rule lately, with IDC pegging
According to research firm Gartner Inc., in 2006 the blade server market saw increases over 2005 for revenue (36.5%) and shipments (33%). "For HP and the server industry as a whole, blades are the fastest-growing segment of the market," said Steve Gillaspy, group manager for HP BladeSystem.
Industry leaders HP and IBM (as well as Dell Inc. and Sun Microsystems) are aggressively pursuing the blade market by touting the inherent manageability of blades in conjunction with cost and power savings. Offering more memory, smaller footprints, and improved power and cooling efficiencies, blades are being positioned as the solution to the woes that plague data centers, namely space constraints and power consumption. With improved memory and I/O, blades work well with virtualization, enabling data center managers to further consolidate physical servers.
Despite blades' technical and market gains, however, data center managers continue to have misgivings about the technology; chief among them are up-front cost and ongoing cooling efforts. In TechTarget's 2007 Server Decisions Survey, nearly 30% of 218 respondents said that cooling issues were enough of a concern to prevent their organization from investing in blade servers. About 27% of respondents cited the cost of the chassis as a concern, while some 22% reported that blades weren't appropriate for organizational workloads. Topping out the list of concerns were the vendor lock-in associated with buying a chassis (21%) and a lack of flexibility in terms of peripherals (21%).
While many data center managers still have concerns about blades, the lingering question remains: Are they a viable alternative to rack servers?
For data center managers, the answer to this question is likely a multifaceted one involving costs, technical needs, and architectural and infrastructural requirements. But one place to start is to evaluate the current technical aspects of blades.
When the first generation of blade servers hit the commercial market in 2000, the technical specs of the new form factor didn't stack up against traditional rack-mount servers. While blades always had an advantage in terms of density -- individual servers are stored vertically in pre-wired and pre-configured chassis -- the I/O, CPU and memory capabilities of individual servers were limited when compared head to head with individual rack servers. While it was a major selling point for footprint-strapped data centers, blades' higher density involved a significant tradeoff: The heat generated by densely packed components often presented cooling challenges that outweighed the floor-space savings.
"When blades were in their infancy, reliability was low, with failovers occurring a lot more than with rack servers," said Dave Leonard, CTO of Infocrossing Inc., an IT outsourcing provider. "When the internal components failed in the chassis of an early blade system, the only way to re-set it was with a paper clip. … That's the level of maturity that blades had."
As their increasing popularity suggests, blades have come a long way. In terms of technical features, typical blades now include dual- and quad-core processors with up to 64 GB of memory and multiple peripheral component interconnect slots; in essence, they feature many of the same options as their rack-mount counterparts. First-generation blades were low-powered machines that worked fine as Web servers, according to Barb Goldworm, a consultant and the author of Blade Servers and Virtualization: Transforming Enterprise Computing While Cutting Costs. Today's blade servers are more reliable and can handle many mission-critical processing workloads; thus they present a viable alternative for data centers that want to replace rack-mount servers, Goldworm said.
The decision about whether to go with blades ultimately hinges on customer preference for one server form factor over another. As for Infocrossing, Leonard said that his company is now relatively impartial to blade versus rack-mount servers -- not a ringing endorsement by any means, but progress nevertheless from the days when system administrators protected themselves from a server failure with a paper clip.
But data center managers still have tradeoffs to consider. On the surface, blades may appear more expensive than rack-mount servers because of the cost of the chassis that is required with blades. Yet the initial capital outlay with blades can be offset with savings down the road, provided a data center is equipped with adequate cooling and power capabilities.
On the other hand, basing a blade purchase primarily on saving square footage -- and thus money from a facilities perspective -- may end up taxing power and cooling if the facility isn't designed to accommodate the higher-density characteristics of blades. And data center managers must also recognize workload considerations; while blades may make sense fiscally and facilities-wise, rack-mount servers may be a better option for highly demanding workloads.
The economic argument for blade servers
As the technical gap between blades and rack-mount servers continues to close, data center managers need to consider not only saved floor space and higher initial cost but also the particularities of their environment.
In terms of capital investment alone, buying blades doesn't make economic sense for organizations that need only a handful of servers, since they have to figure in the cost of the chassis. But according to examples from HP, the cost of blades can break even with rack servers once an organization reaches three servers to eight servers (based on configuration). Assuming $5,697 for an HP c7000 enclosure plus Ethernet and Fibre Channel switches, the break-even point for HP BL460c blade servers (with a list price of $6,224) is the equivalent of approximately four HP DL360 servers (with a list price of $7,152). On the higher end, seven BL460c blade servers with SANs using Pass-Thrus cost $65,504, compared with $65,457 for an equivalent setup with seven DL360 servers.
According to HP, the more blades you have, the more dramatic the total-cost-of-ownership (TCO) savings. Using the same BL460c blades, HP advertises the TCO for 16 blades at $139,156. The equivalent rack configuration has a TCO of $228,165, a premium of almost 64%.
While there are no standard chassis sizes -- industry leaders IBM and HP have chassis capacities of 14 blades and 16 blades, respectively -- a rule of thumb is that customers need to buy enough servers to fill half the chassis in order for the cost of blades to break even with the cost of rack-mount servers. But because the chassis includes all the components needed for cabling, switching, networking as well as power supplies, customers don't have to buy as much ancillary equipment as they do with rack-mount servers. For organizations that need several servers -- on the order of 10 or more -- blades cost less than rack-mount servers.
"People normally tend to simply look at the initial bottom line," said David Marshall, director of business development and product management at InovaWave Inc., a provider of virtualization performance software, and the owner/operator of VMBlog.com, which covers blades and virtualization.
"In a greenfield environment, in which data centers are looking to add to capacity, the key to comparing acquisition costs between blade servers and rack-mount servers is to keep in mind all of the additional interconnect components, and the hardware and software required, which come standard with blade servers," Marshall said. "When you start adding in the costs of HBA [host bus adapter] cards, Ethernet ports and cables, power cables and power distribution units [PDUs], KVM cables and switches, and the management of software and hardware redundancy features, the degree of cost separation gets wider," he said.
So the initial outlay for blades -- provided that an organization is buying a critical mass of servers and adding to overall capacity -- can compare favorably with the purchase of rack-mount servers and all the attendant equipment required.
The economic rationale for blades versus rack-mount servers is rarely a straightforward exercise of comparing spec sheets. Consider the decision-making process for Richard Siedzik, the director of communications and telecommunications services at Bryant University in Smithfield, R.I., who opted for blades. Siedzik compared configurations on two IBM pSeries standalones with a seven-blade system. Including storage costs, the price differential was between 30% and 35% less with blades than with standalones.
In addition to servers, "we were consolidating all of our storage," Siedzik said. "If we had to buy individual servers, we would have to connect each of them to switch ports on the storage. We're only taking up two ports on our fiber network with 14 blades. With 14 traditional servers, we would have to have 28 ports. That could get very expensive."
As Siedzik's experience indicates, there may be cost savings associated with blades that isn't readily apparent. Once a blade chassis is configured, new servers can be added relatively quickly (provided, of course, that the chassis isn't already filled to capacity), which in turn can reduce the tendency to over-provision servers.
For some blade makers, the ability to combine the machines with management software that automatically distributes workload is a major selling point. "There are a lot of cost savings from dynamic repurposing and high availability that are not intuitive," said Susan Davis, vice president of marketing at blade maker Egenera.
At Egenera, the proprietary software PAN Manager allows systems administrators to pool blades and dynamically distribute workload among them as needed. "We can accommodate the separation of the physical server from any logical idea of what's running on that server," Davis said. One Egenera customer in the financial services industry uses blades during the day for trading applications and then repurposes them with a different operating system to do batch reporting overnight.
As they determine whether blades are more economically attractive than rack-mount servers, data center managers have to consider several factors inherent to their organization's environment. In general, the more servers required, the more likely that blades will save money in terms of floor space and hardware costs. Still, while blades may compare favorably with rack-mount servers on paper, any comparison should account for existing equipment, current workload needs and future requirements. Replacing servers rather than adding capacity, for example, will allow data center managers to re-use existing cables, switches and power supplies, thereby negating some of the cost savings promised by blades. Still, for organizations that want to accommodate future growth without over-provisioning today, blades appear to pay off.
Let us know what you think about the story; email Megan Santosus, Features Writer.