Total Cost of Ownership (TCO) Analysis Linux on HP BladeSystem vs. Sun Solaris and x86 Rack-Mounted Servers Robert Frances Group July 2005 HP sponsored this study and analysis. This document exclusively reflects the analysis and opinions of the Robert Frances Group (RFG) author, who has final control of its content. 120 Post Road West, Suite 201 Westport, CT 06880 http://www.rfgonline.com Executive Summary The Robert Frances Group’s total cost of ownership (TCO) calculation measures four drivers of cost – including hardware, software, labor and overhead – and factors in growth in the number of server infrastructure from deployment to one year after deployment, and then a static number of devices thereafter. The Linux on HP BladeSystem solution had the lowest TCO in all cost categories, with hardware acquisition providing the largest cost advantage. After three years, the TCO for an 8-server blade Linux on HP BladeSystem (Blade rack) solution was found to be $185,770, 9.4 percent lower than a comparable x86 rack-mounted solutions (x86 rack) and 16.8 percent lower than a Sun Solaris rack-mounted servers (Sun rack). In the area of server hardware costs, the study found the Linux on HP BladeSystem solution to have the lowest acquisition cost per 8-server blade configuration throughout the model’s timeframe. At initial deployment, its cost is 6.2 percent lower than the equivalent x86 rack solution, and 30 percent lower than a Sun rack-based solution. After 3 years, with the solution fully built out to 250 servers, the HP BladeSystem cost advantage rises to 13.8 percent vs. x86 rack and 29.2 percent vs. Sun rack. Overall 3-Year TCO Components of 3-Year TCO Linux on HP BladeSystem x86 Rack-Mounted Solaris on SPARC $2,082,618 $2,314,434 $2,680,299 Hardware $883,266 $1,368,737 $927,429 $945,094 SW $1,661,091 $1,714,219 $1,836,555 $1,672,358 Labor $1,908,387 Overhead Software costs had little discernable impact on differences in TCO across the three platforms examined in the study. For all platforms, software costs were tabulated to be approximately 15 percent of overall TCO throughout the three-year timeframe but excluded the costs of custom applications. The labor associated with deploying and maintaining a solution is one of the most important drivers of TCO differentials, accounting for an average of 27.1 percent of TCO. The average labor cost (per 8 blades) for the HP BladeSystem solution was 14.4 percent lower than the x86 rack solution and 15.3 percent lower than the equivalent Sun rack solution. After 3 years, the total labor costs for the HP BladeSystem solution were 20.2 percent lower than x86 rack and 18.2 percent lower than Sun rack. One of the key drivers of low labor TCO for Linux on HP BladeSystem is the ease of deploying new systems, defined in the study as provisioning, configuration and integration. The HP BladeSystem advantage in this area is most evident in the first year of deployment, as the solution scales. During this interval, the cost differential for x86 rack rises from 13.4 percent to 17.8 percent. For Sun rack the cost difference is even larger, rising from 14.7 percent to 19.9 percent. In comparing the overhead of HP BladeSystem vs. x86 rack and Sun rack, the findings reinforced previously held assumptions. Looking at overhead as a whole, the 3-year TCOs of both x86 rack and Sun rack exceeded that of HP BladeSystem by 9.6 percent and 13 percent, respectively. The key driver of lower overhead for Linux blade was data center costs. The three-year TCO related to data center costs were30 percent higher for the x86 rack solution and 40 percent higher for the Sun rack solution versus the HP BladeSystem solution. Copyright © Robert Frances Group, Inc. All rights reserved. 2 Goals and Methodology White Paper Goals HP engaged the Robert Francis Group to perform an analysis comparing the TCO of a Linuxbased solution running on an HP BladeSystem platform to that of similar solutions running on: • x86 rack-mounted servers with a mix of industry standard operating systems • Sun Solaris on Sun SPARC rack-mounted servers Within the study, TCO is defined broadly to include the server hardware itself, as well as all software, labor, storage and associated overhead costs. Each of these cost categories are in turn broken down into their constituent elements (e.g., software types). The goal of the analysis is to understand the cost dynamics of these environments along two dimensions. First, how do costs vary over a 3-year ownership interval, holding the scale of the solutions constant? Second, how do these costs compare as the solution is scaled up over its first 12 months? This dual approach was used to underscore the sensitivity of blade TCO to changes in the solution scale that would not be apparent if a “constant number of servers” approach was solely employed. Sample selection and data gathering Of the 10 companies RFG interviewed, six ran Linux on HP BladeSystem environments, two ran x86 rack-mount servers and two ran Solaris-SPARC rack-mount servers. To expand the number of data points in the study, the six Linux on HP BladeSystem respondents were also asked for their expected hypothetical costs if they had deployed on the x86 and Solaris-SPARC rack environments, respectively. While respondent size and industry were not used as selection criteria, the majority of respondents would be classified as service-oriented companies, with Linux on HP BladeSystem respondents concentrated among large banks, insurers and financial service providers. This generally reflects the earliest adoption of blade servers. In addition to the interviews, RFG also leveraged its extensive data on Sun and x86 implementations drawn from approximately 25 interviews. Data for the study was gathered using a two-track approach. To obtain qualitative and select quantitative information, RFG conducted telephone interviews with IT personnel such as architects, distributed computing directors and managers. The majority of quantitative information was obtained through a standalone instrument, administered after the discussion. This made it more feasible for respondents to gather data from multiple sources or forward it to other individuals. System Profiles and Pricing Assumptions For Linux on HP BladeSystem respondents, the core system used was the HP ProLiant BL20p. Within respondents’ installed base, a total of eight configurations were identified, determined by type and number of processors, RAM, internal storage and SAN connectivity. For each customer, the number of servers within each configuration group was tracked. Thus, if a customer had multiple configurations, costs were calculated separately for each group. Server costs were calculated as an expense in the period the purchase was made (depreciation was not calculated or tracked) using two sources. First, customers were asked what they paid on a per-blade basis (as well as the approximate discount off list they received). Second, RFG used HP’s Web-based pricing and configuration tool to generate a list price for each configuration group that could be used for cross-checking purposes. After accounting for discounts, the configured price was Copyright © Robert Frances Group, Inc. All rights reserved. 3 compared to the customer’s stated price. All fell within 3 percent +/-. The final element to be accounted for in calculating blade server costs were shared system resources, such as equipment racks, system enclosures, power distribution, cooling and cabling. Based on discussions with respondents and RFG research, these costs were assumed to be $3,500 per enclosure for the HP ProLiant BL20p. The TCO model also assumed a ratio of 8 blades to one enclosure. As such, the total server cost for each respondent was calculated as: # servers with ( Configuration 1 X Price/blade with Configuration 1 ) + servers with ( #Configuration 1 X ) $3,500 8 This was calculated for each server configuration employed by the respondent. To determine the cost of x86 and Sun rack-mounted systems, the same approach was made, with the obvious exception of enclosures. To determine these costs, RFG needed to identify corresponding systems for x86 and Solaris-SPARC rack-based solutions, respectively. As with pricing, this was determined through a mix of respondent guidance and RFG research. The systems ultimately chosen for comparison were the HP ProLiant DL380 and the Sun Microsystems Sun Fire V240. As with the HP ProLiant BL20p, eight configurations of these systems were identified and priced. Qualitative Findings Overview of Implementations The architectures of solutions covered in the study fell under two respondent-defined categories: • Grid or “grid-like”—In which a cluster of blade servers are controlled by a dispatching server to perform a particular function (e.g., calculating security pricing). This also would include massively parallel processing architectures, in which large amounts of computing power are generated by pulling together a large number of blades into a single entity. With numerous subtle variations, this category represented approximately 80 percent of profiled systems. • Three-tiered—Including tiers for Web serving, Web application services and backend databases. Elements include load-balancing, security/DMZ, LDAP server, etc. Within this architecture, the blade servers were used across all tiers. This represented approximately 20 percent of profiled systems. The number of processors per blade is a key decision made by architects. The pattern seen in the study was that most respondents (about two in three) maintained a strategy of uniformity throughout, while others slightly modified their approach as they developed more experience using blades. These changes were typically downward, with respondents shifting a larger share of their computing to smaller scale systems (e.g., from a mix of four-way and dual processor to Copyright © Robert Frances Group, Inc. All rights reserved. 4 all dual, or from all dual-processor systems to a mix of single and dual. The goal in all cases was to obtain more granularity and flexibility as they scaled (“We didn’t want to overbuild.”). Within the profiled solutions, the majority of deployed applications were complex, proprietary and/or highly customized, reflecting the highly specialized needs of the banking and financial services sectors. Respondents also generally specified that their preference was to focus their blade-based deployments on applications where horizontal scalability was deemed especially important. Examples cited by respondents include: • • • • • • Risk management Transaction processing Customer self-service processing High-volume database access Data mining (generally off the shelf, e.g., Cognos) Document management(e.g., LDMS) Finally, in the area of storage, respondents were found to employ a mix of SAN, attached and internal. Roughly one half of respondents employed internal or attached storage, citing its main focus on maximizing application performance. The other roughly one half of respondents that used SAN storage cited flexibility as the central driver of its storage strategy. Goals: Why Blades? Asked unaided why they selected blade technology in general, respondents typically cited a combination of factors—one primary, one still important but secondary. The most common were: • Optimizing application performance/low cost—Performance: “We are compute intensive. The architecture we chose gave the ability to separate blade servers into different dispatchers, which enabled us to leverage the full capabilities of the application. Cost: Our goal is to minimize our cost per CPU, and the most important thing is simply to have the capacity to add capacity.” Source: Large bank. • Faster deployment/more virtualization—Faster: “We have to be extremely responsive to our internal customers, which means we need to minimize operational bottlenecks—from getting management approval for every server to getting them from the loading dock.” Virtualization: “We're definitely moving in the direction of more virtualization, and blades enhance our ability to do this.” Source: Large financial services provider. • Need for standardization/processing throughput—Standardization: “We had a hodgepodge of systems that needed upgrading. We wanted something that was easy to grow and manage.” Throughput: “Our strategy is to stripe data across many systems – as opposed to striping data across many spindles – to get maximum performance. Blades are ideal for that.” Source: Large consumer information services provider. • Need for standardization/systems management—Standardization: “The situation before was that every project had its own hardware, software stack and systems management infrastructure. The idea is to have numerous standardized commodity units in the infrastructure.” Systems management: “We needed to be sure we had an easy way to set up servers to handle new workloads as they came up.” Source: Large insurance provider. • Faster application deployment/”commoditized” hardware —Faster: “We wanted to have an on-demand environment where we can reduce time-to-market for new applications. Rather than having a 30- to 60-day deployment cycle, we can do it same day.” Commoditized: “Our ultimate goal is to treat hardware as a commodity, and blades did this.” Source: International bank. Copyright © Robert Frances Group, Inc. All rights reserved. 5 • Standardization /ease of management—Standardization: “Having an industry-standard platform makes it easier on our customers and our internal developers. Once it became clear that blade technology was here to stay, it became a no-brainer.” Management: “We get the power we need by pulling together a large number of systems into a single entity; this can include thousands of nodes in the single system. With blades, we saw tremendous savings in cabling as well an ability to manage them as a single entity.” Source: European information services provider. TCO Findings Overall TCO As discussed, RFG’s TCO calculation factors in growth in the number of servers from deployment to one year after deployment, and then a static number of servers thereafter. For the average implementation, the number of servers rose from 120 at deployment, to 170 after 6 months to 250 servers at year 1—and held at that level through years 2 and 3. As Exhibit 1 shows, Linux on HP BladeSystem TCO is the lowest throughout the first three years, followed by x86 rack-mounted servers and SPARC on Solaris rack-mounted servers. At deployment, the gap is smallest between Linux blade and x86 rack. In the first year, as the number of servers grows, this gap grows modestly, while the TCO difference between Sun rack and x86 rack narrows modestly. In the last two years of the model, the TCO for HP BladeSystem diverges from x86 rack and Sun rack at about the same rate. After three years, the TCO for an 8-blade HP BladeSystem solution was found to be $185,770, 9.4 percent lower than x86 rack ($205,033) and 16.8 percent lower than Sun rack ($223,273). As Exhibit 2 on the following page shows, the overall cost of an HP BladeSystem solution rises from $2.5 million at deployment (with 120 servers) to just Exhibit 1 Overall 3-Year TCO (per 8 Blade/8 Server Configuration) $225,000 $175,000 $125,000 At Deployment 6 Months 1 Year 2 Years 3 Years Linux on HP BladeSystem $166,658 $152,870 $126,080 $155,861 $185,770 x86 Rack-Mounted $175,558 $164,557 $137,218 $171,036 $205,033 Solaris on SPARC $199,524 $186,432 $154,328 $188,737 $223,273 120 170 250 250 250 # of Servers Deployed Source: RFG 2005 Copyright © Robert Frances Group, Inc. All rights reserved. 6 Exhibit 2 Overall 3-Year TCO (for Full Solution)/Components of 3-Year TCO $7,500,000 $5,000,000 $2,500,000 At Deployment 6 Months 1 Year 2 Years 3 Years Linux on HP BladeSystem $2,504,031 $3,259,954 $4,069,238 $5,030,424 $5,995,711 x86 Rack-Mounted $2,713,160 $3,509,182 $4,603,896 $5,695,396 $6,792,637 Solaris on SPARC $2,997,846 $3,975,662 $4,980,931 $6,091,484 $7,206,138 120 170 250 250 250 # of Servers Deployed Linux on HP BladeSystem x86 Rack-Mounted Solaris on SPARC $2,082,618 $2,314,434 $2,680,299 Hardware $883,266 $1,368,737 $927,429 $945,094 SW $1,661,091 $1,714,219 $1,836,555 $1,672,358 Labor $1,908,387 Overhead Source: RFG 2005 under $6 million after 3 years (with 250 servers). In absolute dollar terms, the TCO gap for x86 rack rises from $200,000 to nearly $800,000 after three years. By comparison, the Sun rack TCO differential grows from almost $500,000 at start-up to $1.2 million. As the lower portion of Exhibit 2 shows, the HP BladeSystem solution had the lowest TCO in all cost categories, with hardware acquisition providing the largest cost advantage. The following sections explore the underlying cost factors driving these TCO differentials in greater detail. Servers In the area of server hardware costs, the study found the HP BladeSystem solution to have the lowest acquisition cost per 8-blade configuration (see Exhibit 3) throughout the model’s timeframe. At initial deployment, its cost is 6.2 percent lower than the equivalent x86 rack solution, and 30 percent lower than a Sun rack-based solution. After 3 years, and with the solution fully built out to 250 servers, the HP BladeSystem cost advantage improves to 13.8 percent vs. x86 rack but the advantage remains roughly the same vs. Sun rack (29.2 percent). The study found that for the HP BladeSystem solution, the cost of shared system resources (enclosures, etc.) are equal to or less than x86 and Sun rack depending on the network and storage infrastructure deployed. Copyright © Robert Frances Group, Inc. All rights reserved. 7 Exhibit 3 3-Year TCO: Server Hardware (per 8 Blade/8 Server Configuration) $70,000 $65,000 $60,000 $55,000 $50,000 $45,000 $40,000 At Deployment 6 Months 1 Year 2 Years 3 Years Linux on HP BladeSystem $45,226 $44,389 $44,953 $44,953 $44,953 x86 Rack-Mounted $48,202 $49,948 $52,136 $52,136 $52,136 Solaris on SPARC $65,000 $64,944 $63,472 $63,472 $63,472 120 170 250 250 250 # of Servers Deployed Source: RFG 2005 Software Software costs had relatively little impact on differences in TCO across the three platforms examined in the study. For all platforms, software costs were tabulated to be approximately 15 percent of overall TCO throughout the three-year timeframe. Within this general category, the largest subcomponents were systems management software (~60 percent) and operating system software (~25% for Linux blade and x86 rack, but zero for Sun rack). For the solutions profiled, third-party software accounted for a very low share of software costs. This in large part reflected the heavy use of proprietary and/or highly customized software by the broadly defined financial sector, to which most respondents belonged. For a number of respondents, the total investments in highly customized applications ran into the millions. This presented two problems. First, there was considerable uncertainty as to whether these costs should be classified as “software” or “labor,” since the development was performed using internal staff. Second, the sheer magnitude of these costs threatened to distort the TCO results by making software appear to be a more important determinant of TCO than it is in reality. To circumvent these problems, the study excluded the costs of custom applications. Labor Labor associated with deploying and maintaining a solution is one of the most important drivers of TCO differentials. As Exhibit 4 shows, the average labor cost (per 8 blades) for the HP BladeSystem solution was 14.4 percent lower than the x86 rack solution and 15.3 percent lower Copyright © Robert Frances Group, Inc. All rights reserved. 8 Exhibit 4 3-Year TCO: Labor Component (per 8 Blade/8 Server Configuration) $60,000 $55,000 $50,000 $45,000 $40,000 $35,000 $30,000 $25,000 At Deployment 6 Months 1 Year 2 Years 3 Years Linux on HP BladeSystem $46,917 $38,707 $27,293 $34,787 $42,409 x86 Rack-Mounted $54,780 $46,974 $34,009 $43,472 $53,113 Solaris on SPARC $55,356 $47,732 $33,271 $42,480 $51,816 120 170 250 250 250 # of Servers Deployed Source: RFG 2005 than the equivalent Sun rack solution. As the number of servers deployed grew through year 1, labor cost per server fell for each category. From that point, the labor costs for the x86 and Sun rack solutions grew proportionally faster than the HP BladeSystem solution. After 3 years, the total labor costs for the HP BladeSystem solution were 20.2 percent lower than x86 rack and 18.2 percent lower than Sun rack. One of the key drivers of low labor TCO for Linux blade is the ease of deploying new systems, defined in the study as provisioning, configuration and integration. As Exhibit 5A on the following page shows, the HP BladeSystem advantage in this area is most evident in the first year of deployment, as the solution scales. During this interval, the cost differential for x86 rack rises from 13.4 percent to 17.8 percent. For Sun rack the cost difference is even larger, rising from 14.7 percent to 19.9 percent. Respondents pointed to two main factors as having the biggest impact on HP BladeSystem TCO: • “Adaptive” Availability—A number of respondents pointed to HP’s Instant Capacity (iCAP) program—in which preconfigured server blades are delivered and activated only when needed—as providing major savings in acquisition and deployment. “Say we buy three chassis full of blades—24 blades. We order it and get it all on the data center floor in one fell swoop. If I was to buy them sequentially, that would be 24 trips to the loading dock, 24 approvals, 24 cabling requests, etc. The fact that we only have to do this once with blades cut our turnaround time to a day and produces substantial operational savings.” • Cabling Efficiency—A reduction in cabling complexity was cited universally as a major benefit of the HP BladeSystem solution. The core of this benefit is the reliance on the server blade backplane instead of discrete cabling, which enables resources to be reallocated without the need to physically change cabling (as is the case with rackmounted servers). “Our biggest savings is reduced cabling infrastructure. This includes not just the labor of putting this cabling into place but also the ongoing maintenance required.” Copyright © Robert Frances Group, Inc. All rights reserved. 9 Exhibit 5A Labor Costs: Provisioning, Configuration & Integration (per 8 Blade/8 Server Configuration) $40,000 Linux Blade x86 Rack Sun Rack $35,000 At Deployment 6 Months 1 Year 2 Years $21,520 $20,988 $17,257 $21,520 $20,988 $21,520 $20,988 $17,257 $10,000 $17,257 $15,000 $29,732 $24,425 $28,974 $38,132 $20,000 $37,556 $25,000 $32,553 $30,000 3 Years Exhibit 5B Labor Costs: Systems Management (per 8 Blade/8 Server Configuration) $30,000 Linux Blade x86 Rack Sun Rack $25,000 $0 At Deployment 6 Months 1 Year 2 Years $25,054 $20,577 $17,465 $13,718 $8,732 $8,351 $6,859 $10,643 $9,868 $8,546 $11,422 $10,812 $5,000 $9,667 $10,000 $16,702 $15,000 $26,197 $20,000 3 Years Exhibit 5C Labor Costs: Break-fix & Support (per 8 Blade/8 Server Configuration) $9,000 Linux Blade x86 Rack Sun Rack $8,000 $7,000 At Deployment 6 Months 1 Year 2 Years 3 Years Source: RFG 2005 Copyright © Robert Frances Group, Inc. All rights reserved. 10 $4,804 $6,084 $4,575 $4,003 $5,070 $3,812 $3,336 $2,000 $4,225 $8,131 $6,412 $5,736 $3,177 $3,000 $5,801 $4,000 $4,697 $5,000 $7,357 $6,000 Another driver of lower labor-related TCO for Linux blade solutions was systems management, which would include the labor associated with infrastructure and application monitoring, reallocation of capacity, change and patch management, automated workflow provisioning, and remote management. As Exhibit 5B shows, reported systems management costs for x86 rack were 18 percent higher than Linux blade after 3 years. Respondents saw the cost of Sun rack systems management as being 21.5 percent higher than Linux blade after 3 years. Specific examples of the systems management benefits cited include: • More Flexible Resource Allocation—A large bank uses HP BladeSystem servers in a series of grids that run back-office applications. By using blades, the respondent can share capacity across different grids (and grid dispatchers) more efficiently. This in turn has enabled a major improvement in application performance. “Now we can use any free blade for any batch process or compute intensive process. That's the biggest gain. Before they had computations running on 180 servers overnight, now they are able to cut back down to 130. It shortens batch time by two to three hours.” • Cable Management Efficiency—A large provider of credit information uses blades in a grid on which it runs extremely high volume database applications. “We had spent a ton of money on cable management alone. Blades enabled us to achieve big operational savings by making our wire management much simpler.” • Software Deployment—A large insurance provider that uses blades in a grid-based risk management solution cited efficient software distribution as a major benefit. “Under our old [fragmented] systems, we were spending too much of our resources on things like patch and change management. Combined with the standardization of our systems, we expect the tools available under the new [blade] infrastructure to streamline [patch and change management], and probably cut that portion of our systems management in half.” • General Efficiency—One respondent, a large international bank, commented that “Linux systems administration efficiency is five times as efficient as UNIX-based. Now we are managing 100 servers per administrator, compared to 20 under UNIX.” System repair and replacement (i.e., break-fix) was also a significant source of TCO advantage for HP BladeSystem solutions. As seen in Exhibit 5C (on the previous page), the TCO delta of HP BladeSystem vs. x86 and Sun rack solutions grows steadily in the three years from deployment. For Sun rack, the total three-year costs related to break-fix are 4.8 percent higher than Linux blade. Break-fix costs for x86 rack, by comparison, are 24.9 percent higher. The importance of this benefit of HP BladeSystem was clearly evident from RFG’s discussions. The most immediate—and most frequently cited—benefit was the extremely short time required to replace individual blades when they go down. Because failed blades can be replaced modularly, with no cabling issues, the time required is negligible. Though cited less frequently, the software side of replacement—mainly self-configuration, under which a new blade automatically seeks out the management server and downloads the software to configure itself— was also cited as a benefit. Overhead In the context of the study, overhead was defined to include the cost of data center real state (i.e., “floor tiles”), power, cooling and security/fire control. As a proportion, overhead as a share of TCO nearly doubled from year 1 to year 3 (to between 26 and 28 percent for all environments). In comparing HP BladeSystem to x86 rack and Sun rack, the findings reinforced previously held assumptions. Looking at Overhead as a whole, the 3-year TCOs of both x86 rack and Sun rack exceeded that of HP BladeSystem by 9.6 percent and 13 percent, respectively (see Exhibit 6). The key driver of lower overhead for Linux blade was a data center cost. The three-year Copyright © Robert Frances Group, Inc. All rights reserved. 11 Exhibit 6 3-Year TCO: Overhead Component (per 8 Blade/8 Server Configuration) $60,000 $50,000 $40,000 $30,000 $20,000 $10,000 At Deployment 6 Months 1 Year 2 Years 3 Years Linux on HP BladeSystem $24,615 $22,417 $17,156 $34,311 $51,467 x86 Rack-Mounted $26,526 $24,372 $18,968 $37,936 $56,903 Solaris on SPARC $27,631 $24,850 $19,710 $39,419 $59,129 120 170 250 250 250 # of Servers Deployed Source: RFG 2005 TCO related to data center costs was 30 percent higher for the x86 rack solution and 40 percent higher for the Sun rack solution than Linux on HP BladeSystem solutions. This of course relates to the higher levels of density achievable through blade server solutions (estimated by one respondent as being roughly 5 to 10 times that of rack-mounted servers). This attribute factored into many respondents’ decisions to implement a blade solution, but was generally not the deciding factor. Conclusions One of the clear sentiments expressed by respondents was the idea that blade server technology: 1.) is mature enough to handle most production applications 2.) is “the next step in server evolution” in terms of the ability to replace failed systems quickly and 3.) is one of the best options for companies seeking to consolidate, standardize and/or “virtualize” their architectures. In looking at the TCO of HP BladeSystem vs. x86 and Sun rack, there was a relatively clear consensus that the HP BladeSystem had the lowest over a 3-year ownership period. The other point of broad agreement was the idea that blades’ biggest cost advantages were operational and labor-related—not in the cost of the systems themselves. At the root of this was the undeniable benefit of cabling simplicity, which had ramifications both for upfront provisioning as well as ongoing system maintenance and repair. RFG also concludes that while a good deal of blade servers’ widely perceived benefit of having lower overhead was well-founded, some findings were surprising. To some extent, respondents were affected by their companies’ specific situations. For example, respondents with highly constrained data center space and large number of systems tended to emphasize data center cost reductions enabled by blade density. Others—both large and medium-sized—saw strong appeal and cost savings from the ability to scale horizontally in a highly granular way. HP’s iCAP Copyright © Robert Frances Group, Inc. All rights reserved. 12 program dovetailed strongly with this desire, as well as the need to shorten and streamline the deployment of new systems in response to growing resource needs. For blade technology, the issue of cooling is considered important to respondents, who realize that the need to accommodate cooling requirements of new processor and memory technology will enter into their data center space planning and possibly affect server density regardless of the form-factor deployed. Copyright © Robert Frances Group, Inc. All rights reserved. 13