How I Conduct Data Center Site Selection
11 min read
Jeff August, UpStack’s Chief Strategy Officer, details his proven method for evaluating data center facilities using the lessons he learned sourcing colocation and network during his time at Facebook, Dropbox, and Square.
Data center site selection is essentially a real estate transaction and, like any real estate transaction, the specifics of a given project are going to influence the details of the process used to evaluate potential solutions. Building edge nodes for an application or content delivery network will require a different set of priorities than building a large scale deployment to handle the compute functions of thousands of backend servers that support mission critical services.
That said, there are strategic evaluation categories that can be applied to just about any DC project. Once you know what your needs will be for each of these evaluation categories, it is also helpful to apply a dynamic scoring system that compares each of your shortlisted sites so that you can make an “apples to apples” comparison. Using a quantitative scoring system is important to enforce objectivity in the decision making process. Without a system of weighting imperatives and business goals, I’ve seen too many data center projects hindered by “analysis paralysis” and groupthink.
Another thing to remember in this process is that data center site selection is a program, rather than a project. What I mean by this is that site selection is ongoing and made up of many smaller projects. For example, before choosing between data center facilities within a market, you must choose which market is right for you. Let’s assume that you’ve already made it over this first hurdle and you know where you need a facility. Now what?
Without further ado, here is Part 1 of the Jeff August Method for addressing data center site selection. We’ll start with strategic evaluation categories. Part 2 covers the weighted scoring system. Like any process, there is always room to improve, but this one reflects the lessons I have learned (sometimes by screwing it up!) at startups that have become today’s major tech powerhouses.
Once you have defined which market or markets to attack, you have to find providers, structure your requirements and request pricing from vendors in each market. That’s the easy part. After you have received proposals from colocation providers, the next step is to choose who to shortlist among the facilities and providers in each location. In my site selection method, there are seven evaluation categories for you to focus on and gather information about. They are: Cost, Connectivity, Security, Logistics, Growth, Performance and Innovation. I will define each imperative and explain what to look for in greater detail below, but I want to call out some details about connectivity and performance here. Since most modern deployments are hybrid cloud, I break out connectivity as two distinct categories in the matrix – interconnectivity and physical network. Similarly, I break out performance into the sub-categories of availability and reliability.
Let’s look at each one of these strategic categories individually.
Cost is not as simple as the number you see from a data center provider’s quote. While most proposals from a colocation provider will include pricing for space and power with a $/kW and $/Sq Ft value, the longer term cost drivers will need to be accounted for here as well. If you choose a proposal with metered power, you run the risk of future energy cost increases. But you gain the advantage of a lower committed cost if you aren’t going to utilize all available power.
It is also important to consider questions like:
- What does each cross connect cost in the facility?
- What will it cost to use remote hands services and a provider engineer to conduct a card swap for you?
- Can you buy these hours in bulk?
Even the basic questions can affect your total cost of ownership (TCO). Make sure you take a holistic view of how you intend to use the colocation providers slate of services and account for them in your cost calculation.
Sticking with the Vegas theme, you need to go beyond the hotel room rate and price out transportation, concerts and gaming to know what your Vegas trip will truly cost. For your colocation deployment, you will need to price out connectivity and services since the whole point is to have access to computation and getting your info to where it needs to be.
As I mentioned, I break the connectivity category into two distinct areas and score them separately. The first is “physical networking,” which is things like Dark Fiber and DWDM Wavelength Services that are used to tie this specific deployment to the rest of your footprint. You can think of this as Layer 1 services as described in the OSI Model. The other is “interconnection,” which is composed of Layer 2 and 3 technologies that allow for services like IX (internet exchange) Peering, IP Transit, and direct connections to the largest public cloud providers.
To simplify, you will want to know which network service providers have a presence in the building in which you are about to deploy your infrastructure. You will want to know if there is IX Peering in the facility. If there is IX peering, you need to know how many networks have a presence in that exchange and how much of the traffic you exchange can be moved closer to your mutual customers. Also, do you use Alibaba, AWS, Azure or any of the other public cloud service providers? You will want to understand if you can connect to them directly within the facility.
This is all about who is in the building! Back to my Vegas analogy: Is there a Guy Fieri joint slinging burgers off the gaming floor? Maybe Barry Manilow is playing in the hotel’s concert venue? Which is going to give you the better experience or bang for the buck?
An easy way to think about this is “How many barriers exist between my computers and the Ninja Death Squad that wants to steal the information on them?” When you visit a DC you will want to note if it has a gated entry to the parking lot, a man trap as you enter the lobby, a guy behind a desk who checks your credentials, and another man trap before you hit the data center floor.
Other security items you will want to take note of:
- The number of cameras you see as you walk the grounds – if you don’t see them, ask.
- The materials the cages are made out of – can you see into them?
- How many biometric check points did you have to pass through?
- Do cages extend below the raised floor?
- Are the cages covered?
I wrote about the Ninja Death Squad in jest but it is not far from the truth. You will want to approach the facility as if you are a member of a hostile group planning a raid. Ultimately, you need to ask yourself how easy or hard it would be to get to your cage and or locking cabinet.
This category covers the ease of your ability to operate in someone else’s building. How many people are on staff at any given moment? How do you reach out to them if you need something done? How do you give access to your teammates? In these days where folks are ordering fully populated 48U racks, the importance of a shipping and receiving area that can accommodate those kind of racks and the trucks that bring them can’t be understated.
Think of this as Vegas resort-style living for your computers. You want your experience to be as simple and frictionless as possible.
However many racks you are deploying in the space today, I can almost guarantee you will be deploying more in the future. It isn’t enough to have a Right of First Refusal on the space adjacent to you cabinet, cage or suite. You will want to understand how much space, power and cooling are available now and how the data center operator plans to accommodate additional capacity in the future.
- Do they own the land next door?
- Is there uncommissioned space in the current facility?
- Is there plenty of commissioned space with no tenant?
- Can they put your deployment in a part of the building that was recently commissioned?
Forklifting a deployment to another facility isn’t fun or easy. It is best to know that when you plant a flag, your kingdom can grow if and when it needs to.
This strategic imperative warrants a post on its own, but for now we are going to focus on two important performance elements to consider – reliability and availability. In short,
Reliability is how a data center was designed to perform.
Availability is how that data center is actually performing.
For greater detail on performance, you can read this excellent paper pulled together and published by Infrastructure Masons.
Reliability can be expressed in simple terms related to redundancy, like N+1 or 2N+1. Knowing the redundancy model is important, but how important it is for your needs is determined by the specific project for which you are sourcing a data center. Are you sourcing space and power for an edge node to deliver content? Is it okay if that edge node is unreachable for some period of time? If so, high levels of redundancy may be overkill. Knowing the answers to these questions will let you know if a data center in question is designed to meet your needs.
Availability is pretty straight forward – it means the quality of being able to be used. You will want to understand the number and duration of availability-impacting events at a data center you are considering.
Previous Power Failures + Thermal Overload Events + Network Failures (that impacted actual customer services)
The combined duration (in seconds) of those events measured against the total number of seconds in a calendar year will give you a clear uptime metric to consider. For example, any combination of incidents that result in 316 seconds or more will result in a Data Center that has less than the vaunted “five nines” standard of 99.999% uptime. These are things you will want to ask your potential data center partner about early on in your research process .
Most of what I have written about so far should be easy to look at objectively. You can count the number of networks in a facility, or the number of cameras. You can measure the height of a freight elevator and know if your typical rack could be rolled off a giant truck fully populated and moved directly to your cage. Innovation is a little different.
Innovation is going to mean something different to everyone involved in vetting a DC deployment. Is it power efficiency measured in PUE? Is it the exact system of redundancy they use when delivering power? Security advantages or cooling technology? Is it something else?
The answer is in the eye of the beholder, but knowing this and defining priorities with your team will help you to avoid groupthink and to have real discussions about how you expect your infrastructure strategies to evolve in the future.
So, there you have it. These are the strategic evaluation categories that I have found help me compare data centers in a given market: cost, connectivity, security, logistics, growth, performance, and innovation. How you weight each and how important they are to your goals will depend on the specific project, but every project can be evaluated methodically once the weighting is defined. I believe it’s important for businesses to have clear insight into evaluation techniques.
As I mentioned previously, the first step is narrowing down the geographies and data center providers – I wish UpStack existed when I worked at Dropbox, Facebook and Square. The data center sourcing marketplace allows you to search globally, request pricing from hundreds of providers and is designed to provide you with the information you need to help you make better business decisions on a standardized basis. With UpStack, level-set quotes and all-in cost of ownership information is stacked up next to specifications and information about available providers and security. As a matter of fact, with UpStack, all seven of my evaluation categories are within easy reach when evaluating supplier options.
UpStack is vendor-neutral data center marketplace that is free for buyers – we are paid by whatever provider you choose, so our only concern is finding you the best solution for your needs. Our team has over 70 years of combined data center experience, so contact us to see how we can help with your next deployment.
Next, I will discuss how I apply dynamic scoring to these strategic imperatives and how it will help you to choose the best data center facility for your company.