As IT services become an increasingly critical part of everyday business activities, data center downtime (whether in the company’s own data center or in data centers that provide hosting, colocation or cloud services) becomes less tolerable. For companies that rely heavily on their IT networks to perform critical business functions, such as interacting with customers, even a few minutes of downtime can incur tremendous penalty costs. But availability, like any product feature, has a cost, and the better that feature, naturally the higher the cost.
How Much Availability Do You Need?
Obviously, there is no one size fits all answer to this question. Everybody would like a data center that is 100% available (that is, all IT services are available all the time), but not everybody “needs” 100% availability—in other words, sometimes the cost of higher availability far outweighs the benefits. The first step to determining a sensible availability level for your data center is to carefully and honestly evaluate your business needs. This process involves estimated cost analyses and characterization of the criticality of your IT systems. Dr. Mickey Zandi, managing director at SunGard Availability Services, notes the importance of the latter in determining your required level of data center availability: “Companies should look at the criticality of their applications, systems and services that are essential to running the business. This is determined by a required or mandated recovery that is driven by the resiliency of the mission-critical applications.”
Part of a thorough evaluation process is calculating how much money you will lose for each hour (or minute, or another time increment of your choice) of downtime. Costs in this regard include lost business with customers (both short term and long term), employee time diverted from other tasks to get the IT systems running again, employee overtime expenses (if applicable), the value of any lost data, emergency maintenance fees (particularly if the outage occurs during off hours) and additional repair costs that may go on even after service has been restored. Needless to say, you must estimate many of these costs, as they will vary depending on when the downtime event occurs, which systems are at fault and what measures are required to get the facility operating again. But even a rough guess in this area will be extremely helpful when you’re deciding your required level of availability. According to an ADC white paper (“The Three Principles of Data Center Infrastructure Design”), “Uninterrupted service and continuous access are critical to the daily operation and productivity of your business. With downtime translating directly to loss of income, data centers must be designed for redundant, fail-safe reliability and availability. Depending on the business, downtime can cost anywhere from $50K to over $6 million per hour.” Essentially regardless of your position on that scale (albeit some businesses may fall below the $50,000 per hour mark), an hour of downtime represents a tremendous amount of money that could have been invested at the start in infrastructure to improve availability.
Once you’ve estimated the cost of downtime in terms of some time unit, you can compare downtime costs to amortized capital costs and operational expenses associated with data center designs for various levels of availability. When downtime expenses for a given availability greatly exceed the capital and operational expenses for such a design, your best bet is likely to step up to a higher availability level. Eventually, as you go up the availability scale, you’ll reach a point at which the amortized expenses associated with avoiding downtime exceed the cost of downtime itself. At this point of diminishing returns, greater availability can actually harm your business from a fiscal perspective. Assuming you’ve accounted for all short- and long-term downtime costs (a difficult task, particularly if you’re trying to get an accurate estimate), however, this general approach should allow you to determine the level of data center availability that will most benefit your company.
As a side note, properly amortizing capital expenses requires you to estimate the lifetime of your data center—meaning everything in it. This may also require taking into account future growth and additional equipment that may be needed as time goes on. Furthermore, existing equipment will likely be taxed more heavily as demand for services increases. Again, accurately estimating the capital expense of a data center can be a difficult exercise, and at a certain point, seeking greater accuracy simply becomes a waste of time. One simple approach is to simply go by some industry-wide estimates. Doing so can simplify your task, albeit at the cost of less accuracy.
Availability: Cost per Square Foot
“The actual data center construction is tied to the level of redundancy that’s being designed within the data center. If you design a fully redundant vs. a partially redundant vs. no redundancy data center—the cost of equipment, the cost of networks, and the cost of construction— everything changes by the availability and resiliency requirements,” said Zandi. An Uptime Institute white paper (“Cost Model: Dollars per kW plus Dollars per Square Foot of Computer Floor”) cites three drivers of data center construction cost: capacity or density of power and cooling, Tier rating, and floor space in the computer room.
If you know roughly how much floor space you require for your data center, you can use a set of industry estimates to calculate the construction costs for a data center with a given level of availability. For instance, an Anixter white paper (“Data Center Design and Infrastructure Estimates”) cites such estimates from the Uptime Institute regarding the cost of data center construction for various tier levels.
Note, however, that Uptime’s Tier rating and a given availability are not necessarily to be equated: a company could feasibly build a data center that fails to meet all of Uptime’s requirements for, say, Tier III, yet still achieve 99.982% availability (the minimum level of availability for achieving Tier III certification by the Uptime Institute). Nevertheless, these numbers provide something of a guideline. (The Uptime Institute’s Tier rating system encompasses four certification levels: Tier I, Tier II, Tier III and Tier IV. More information is available at the Uptime Institute’s website, or in the Data Center Journal article “The Current State of Data Center Tier System.”)
These estimated construction costs are as follows (from the Anixter white paper).
• Tier I data center (99.671% availability—about 28.8 hours of downtime per year): $450 per square foot
• Tier II data center (99.749% availability—about 22.0 hours of downtime per year): $600 per square foot
• Tier III data center (99.982% availability—about 1.6 hours of downtime per year): $900 per square foot
• Tier IV data center (99.995% availability—about 0.4 hours of downtime per year): $1,100 per square foot
These costs, again, are estimates, but they can provide a starting point at least. In a blog post (“Building a Data Center? Lucky You! Here are 10 questions you should be asking yourself before you begin”), Gartner’s Research VP David J. Cappuccio notes the impact of availability level on cost: “Data Centers are generally defined by tier level; which essentially dictate the availability (up-time) goals for the environment...Many companies use The Uptime Institute’s 4 Tier availability guidelines as a good rule of thumb in early design stages. Determining this tier is critical, as upwards of 60% of your capital budget can be determined by your tiering decision.” To reiterate, however, your company may or may not wish to have its data center certified by the Uptime Institute—hence, not all the baggage of a particular Tier level is incumbent on you when you select an associated availability level. Nevertheless, the Uptime Institute’s Tier rating system provides a more holistic view of data center performance beyond a single number (availability).
The numbers above provide a stark contrast in cost for what appears to be a fairly small difference in availability (99.995% for a Tier IV data center versus 99.671% for a Tier I data center). Measured in hours, the difference in availability is somewhat more striking: 0.4 hours (less than 30 minutes) per year for Tier IV versus 28.8 hours (more than a day) for Tier I. A Tier IV data center costs nearly 2.5 times the cost of a Tier I data center per square foot. This price premium means that simply going for high availability simply for the convenience of it (after all, who likes to lose their IT services?) is a non-starter. But if downtime costs you millions of dollars per hour, you may be able to easily justify a step up to a Tier II, Tier III or even Tier IV data center, depending on how the numbers work out.
The approach to evaluating data center availability described above assumes a fairly static situation: demand for IT services remains constant, so the existing infrastructure is not loaded with more IT equipment to support, for instance. But given the current situation in the data center industry, this is an unrealistic assumption. The Data Center Journal noted in June (“Uptime Institute’s Inaugural Data Center Industry Survey”) that according to the Uptime Institute’s inaugural data center industry survey, “about a third (36%) of respondents anticipate a shortage of power, cooling, available space, or some combination thereof in the next year, and nearly two-thirds (60%) expect to upgrade/renovate an existing data center or construct a new data center within the next three years.”
As a result, your approach to determining your required availability level should take this potential growth into account. Of course, the amount of growth your business must accommodate will depend on your industry and other factors, but regardless of the details, one way of accommodating this growth is to take a modular approach to data center construction and expansion. Anthony J. D’Ambrosi, Chief Sales and Marketing Officer at IO, notes the importance of modularity: “Unlike traditional data center construction, the modular approach allows companies to consume only what they require, when they need it. In addition to eliminating overspending and stranded capital, the modular solution takes the best from data center design, reliability and efficiency, packing everything into a standards-based repeatable and operationally optimized technology platform.”
Modularity allows you to “right-size” your data center to meet near-term requirements while still leaving you the option of adding more capacity as you need it. This approach contrasts with building to meet long-term needs—a strategy that has a very high initial capital cost and that is inefficient on a number of fronts. It also contrasts with the more staccato traditional approach of data center construction: build a new facility (or add on to an existing one) when you run out of capacity. (Building new data centers or adding on to existing ones is still required at some point, but in the modular approach, these are not the only times during which capacity is added.)
Zandi also supports the modular approach: “More organizations are moving to service-oriented architectures and needing full redundancy and higher availability. Because of this, organizations need to evaluate how to design a data center in a module fashion and utilize cloud computing as an augment strategy.” Modularity helps reconcile the need for high availability with the growing demands placed on IT resources by spreading capital expenditures over a longer time and by eliminating the need to predict growth far in advance, thus making the entire build-out process more efficient.
A Pitfall to Avoid
If you’re building a small data center (a “computer room” type facility), then estimating your needed availability using some rough industry numbers may be fine. But if your facility will be large and expensive, then grabbing a few numbers off the Internet is not the wisest approach. In such a case, you will want to hire professional help: although this may cost more in the short term (consultancy fees), it may save you from committing multi-million-dollar mistakes when you build your facility. For instance, using the estimates above, if you build a 2,500-square-foot data center at the Tier III level ($900 per square foot) when you only really need Tier I availability ($450 per square foot), you’ve spent well over $1 million more in construction costs.
The Uptime Institute white paper summarizes this pitfall: “It is not appropriate, nor wise, to base the financial success of any major multi-million-dollar project on free information (like that offered in this or other white papers). As a way to avoid these multi-million-dollar mistakes and unnecessarily foreshortened project life cycles, the [Uptime] Institute urges owners to obtain sufficient seed money to select the design and construction team and actually complete the conceptual design.” Zandi likewise notes the importance of bringing professional help into your major data center project: “Outside resources can organize, catalog and validate what’s mission critical, what are the resiliency requirements and how to address it with the availability of the data center itself.”
The point is this: don’t go it alone. When you’re building expensive facilities, don’t skimp. Professional data center design consultancies have already beaten the learning curve, and they can help you avoid common (and expensive) errors in data center planning and construction.
High availability is expensive. As a result, you must carefully evaluate how much availability your data center really needs before embarking on construction of the facility. If you choose a level too high, you may end up paying more in construction costs than you would have lost in downtime had you built a lower-availability facility. The general approach and basic numbers above can help guide you to a rough estimate of cost for building a data center of a specific Tier rating (with its corresponding availability level). Nevertheless, because Tier and availability and not precisely synonymous (although they may be related), the best approach is to consult with data center design professionals, particularly when you’re dealing with a multi-million-dollar project. At this high level, small errors in judgment can still be very expensive. In addition, don’t neglect the importance of modularity in your data center design. A modular approach allows you to combine the benefits of “right-sizing” your data center with the flexibility of being able to quickly add capacity as your business needs change. Thus, you can plan for growth without being forced to pay for it at the start—instead, you can make capital investments as they’re needed. Availability is expensive, but you can still choose the right level of availability that your company can afford.