The pace of technology innovation always seems to be accelerating, meaning the multimillion-dollar investment in a data center can keep you up at night. How can you plan ahead for even the next financial quarter?
Whether forecasting data center demand for a corporate, in-house facility or a wholesale or IaaS operation, you need to accurately predict future needs of internal and customer IT in terms of computing power (CPU/RAM) and storage. That has to be matched with how much power is necessary, the equipment you need, the space to hold it, supporting infrastructure and staff, cooling throughout different seasons, and the fiber circuits to connect it all to the world—among other items.
Laying the Groundwork for Future IT Load
We’re assuming you’ve already picked a location for your data center, which is its own bag of considerations. The first step is to evaluate your existing infrastructure and IT environment. Categorize the workloads and interview users about how they expect systems to function. Measure service level agreements and how they have stacked up over the years (have you maintained availability? What factors contributed to any downtime?) Tally up resource use throughout all facilities and then figure out the ratio of resources to workloads.
If you are entering a new market, you have likely already performed demand projections and market analysis, which can help you estimate the equipment you will need. How many resources do your current customers use, on average? Extrapolate to your new market.
Examine your data center site. Is retrofitting an option? That imposes some limitations on power, cooling, and space—only so much will fit. In some ways this makes your job easier. Even though it is a puzzle to fit your requirements into a set area, limited choices make the decision process smoother.
Planning for Power
Power is the biggest constraint in data center capacity planning. Use your gathered historic power usage, plus plans for equipment purchases, server consolidation, and best estimates for customer acquisition, in the case of colocation facilities.
Based on your estimates, decide the rack density, which can range from 3KW to 40KW per rack. A high density rack is easier to achieve with modern, thin servers and virtualization technology. Based on this rack density, calculate a total number of racks based on your white space footprint. You’ll have some idea of how much power is necessary, but leave some breathing space by planning for at least 2.0 PUE, even if your facility will be more efficient.
Call local utilities and negotiate a favorable, long-term rate. There may be additional taxes and fees on your energy use depending on the country. For example, Europe levies carbon taxes on data centers.
Even though you are planning for extra power, figure out your cooling requirements based on available equipment (after talking to vendors and pulling together a few options). This can make a huge difference in your power draw. Implement cold or hot aisle containment to save some energy, chimney vents, and, if your climate allows, free cooling.
You’ll probably need an A+B redundant power feed to the data center, and some reliable and redundant Power Distribution Units (PDUs) that can handle the appropriate rack loads. There is management and monitoring software available that will record power, temperature, and other statistics to help you plan for fail-proof power supplies.
Modernizing with Cloud and Modular Builds
Modular data centers are one way to save money and avoid overprovisioning mistakes (also allowing yourself time and space to correct any errors in your original demand forecast or infrastructure setup). Basically, with a modular setup, you build out a certain percentage of your available space, and plan to add more generators, cooling units, and power when necessary. A well designed modular setup can be built out in as little as 90 days.
Old-school IT can’t match the efficiency offered by multitenant or private cloud. Virtualization allows less servers to do more work, freeing up rack space. Planning to push heat limits, humidity, and server utilization will help you maximize available space and avoid overprovisioning.
Keep in mind that customers or IT will often overestimate the capacity they need. Take that into account and assume they are telling you peak resource use. While you will need that infrastructure to be available, normal capacity might be much lower.