As application mirroring fully matures and enables synchronous processing, there are some in the industry that believe IT organizations will no longer have a need for the traditional robust, resilient data centers. These are expensive to build due to their redundant on-site power conditioning and generation.

 

These industry professionals propose a radical new approach: replace one highly robust data center building with two low-cost-to-construct data centers that mirror each other but rely on the utility grid for power. At face value, this approach provides the needed redundancy at a lower total cost of ownership, and may be of interest to hyperscalers or service providers with large, widespread operations and low margins. However, examining this alternative and its underlying drivers can help you better understand whether it may play out in your enterprise IT organization in the future.

 

As I see it, the “two vs. one” scenario presents two basic issues. The first is whether the availability of two data centers that rely on utility-only power are really as good?or better?than one highly resilient data center. The second issue is analyzing the financial model used to compute potential savings.

 

Two vs. One

From a pure mathematical approach, each of the two data centers with utility-only service would need to provide 99% availability to equal the same availability as one highly resilient 99.99% available data center. Sounds reasonable, doesn’t it? How many times do you lose power at your house over the course of a year? Did you experience 87.6 hours of power outages?or 1% downtime?last year (8,760 hours)?

 

Probably not, unless it was due to a catastrophic natural event, such as a hurricane or ice storm. It’s more likely you lost power perhaps five to 10 times, with each event lasting from or a few milliseconds to a few hours. Here’s the rub-data centers aren’t like your house. Recovery of a data center requires more than resetting the clock on your oven. It requires people, systems, software and networks to be restarted in a set process. Each outage takes a minimum of four hours to recover and could easily take more than eight hours should something go wrong.

 

With two data centers, you have double the work. You have twice the human error, twice the network issues, twice the application issues, etc. Even if you assume that, from a power perspective, both approaches provide equal availability when the power is restored, these other factors can have a very real impact on your downtime. From this standpoint, a single high-availability data center is clearly the better approach.

 

Cost Savings

Now let’s turn our attention to the financial model. Could the savings outweigh any availability considerations?

 

It is true that two data centers, each without on-site power conditioning and generation, would be much less costly to construct than one high-availability data center. However, we need to factor in that they would also require twice the amount of IT equipment. First, let’s take a closer look at the capital costs on a per-cabinet basis.

 

  Two Utility-only Powered Data Centers One High-Availability Data Center
Estimated $ per kW of capacity $3,000 $20,000
# of cabinets with 10kW power density 2 @ 10kW 1 @ 10kW
Construction cost per cabinet $60,000 $200,000

 

Even with double the number of cabinets, the capital costs for initially equipping the two utility data centers are significantly less-to the tune of $140,000. It is safe to say that the savings would be more than enough to fund the cost of a second set of IT equipment, even taking into account the additional network infrastructure, cabinet costs, etc. The two utility data center option is clearly the winner when looking purely at capital costs.

 

But we must remember that these are not one-time costs. Typically, IT equipment will be refreshed six to 10 times over the 30-year life of the data center. In addition, there are greater operating costs across two facilities (double property tax, double electricity, etc.).

 

The next step is to compare capital equipment costs over time. Let’s be conservative and say that we refresh the IT equipment every five years.

 

  Two Utility-only Powered Data Centers One High-Availability Data Center
Annual depreciation on facility construction (30 years) $2,000 $6,700
OPEX related to second set of IT equipment ($140,000/5) $56,000 $28,000
Total Annual OPEX $58,000 $34,700

 

To equalize both approaches on an annual operating expenses basis, the IT equipment for two data centers would have to cost less than $23,500 per cabinet at a five-year refresh rate. Therefore, the long-term cost for equipping two data centers is much higher. This doesn’t take into account the extra overhead costs associated with a second data center such as networking, electricity, labor, sales and property taxes, etc.

 

Based on these estimates, one high-availability data center holds a clear cost advantage over two utility-only data centers. In fact, when you analyze capital and operating expenses together, one high-availability data center stands out as the best investment in the long term.

 

One + One Doesn’t Necessarily Equal Success

In spite of application mirroring advances, I believe we will continue to see companies choosing high-availability data centers for the foreseeable future. The return on investment clearly does not yet justify running two utility-only data centers unless IT equipment cost per cabinet is less than $23,500, and even then there is no clear advantage from a reliability perspective.

 

Will this change in the future? For hyperscalers with multiple sites, the attraction of two utility-only sites may come sooner than we think. As the cost of IT equipment and power continues to decrease, we may see more service providers embracing lower-cost facility designs and sacrificing some level of availability.

 

The typical enterprise IT organization has a more limited data center portfolio and will likely continue to embrace the resilient data center design approach for the near future. However, as topography moves to processing workloads in parallel paths within a single high-availability data center, we will see some IT shops moving to a less robust facility infrastructure in the short term. I expect to see changes such as data centers that use a single level of on-site power and uninterruptible power supply (UPS), versus the traditional approach that uses highly redundant on-site generators and UPS systems.

 

While high-availability data centers may be the current norm, I believe substantial change is just around the corner for both NetApp and the industry. The big unknown in this scenario is the hybrid cloud. What are the implications and opportunities as enterprise IT more fully leverages the cloud? Will data center footprints shrink and designs be simplified? We’ll explore what these trends mean to NetApp’s data center strategy in future blogs.

 

The NetApp-on-NetApp blog series features advice from subject matter experts from NetApp IT who share their real-world experiences using NetApp’s industry-leading storage solutions to support business goals. Want to view learn more about the program? Visit www.NetAppIT.com.

 

Mark Skiff