Clearly, if you are responsible for your organization’s or client’s IT capacity planning, a lot is riding on your shoulders. Overprovisioning for hardware means your company has to absorb excessive capital and operating costs. On the other hand, if your technology forecast falls short of demand, you put the user experience at risk.
In the past, you may have waded through data in Excel to create graphs when confronted with such issues. But you want to tackle the forecasting task as accurately and efficiently as possible. It’s easier to create a forecast using a tool designed specifically for the purpose — one that collects data on your IT infrastructure usage and includes forecasting models that can extrapolate the data into the future.
Here are just a couple of examples of what you can do with such a tool:
- Project Your Filesystem Usage
Figure 1 shows filesystem usage in our production database. The solid line that goes from February to September 2016 shows actual growth in usage; the dotted line is a projection of how it will build in the future. The dashed line across the top depicts approximately 3.5 terabytes of capacity that are available. If we look at where the two lines intersect, you can see we need to have additional capacity in place by August of 2017.
- Visualize the Future of VMware Environments
While the data behind CPU utilization in VMware environments is less straightforward, with the right tool, you can use it to predict the future. Figure 2 shows VMware hosts in a V center, one represented in purple and the other in orange. The dotted lines that cut through the erratic motions of actual utilization are derived from a regression analysis (remember, back in algebra class, y = mx +b).
For most situations, regression analysis will give a fairly realistic picture of the future. However, if this equation does not fit your data well, it’s good to have other options. For instance, polynomial squares would produce a different picture as shown in Figure 3. The purple projection accelerates more slowly as it moves into the future. The orange line makes an arch, which is probably unrealistic as the data will likely not drop to zero, so is not a good fit. When forecasting, always add your judgment to determine which model works best for your data.
- Zoom in for Better Forecasts
Your human judgment also comes into play when determining how much data to use and from which period. For instance, you may have background information on why utilization increased quickly at some point. Your company might have changed an application or acquired a large customer, affecting utilization rates. So instead of looking back at your entire history, you might want to zero in on the period after the change occurred.
Below shows growth (and decline) that the model predicts when we zoom in on the data depicted in Figure 2 from the middle of November to the beginning of December 2016. The purple dotted line grows more rapidly based on this subset of the data, reaching 100 percent by March 2017. This is quite different from the 80 percent utilization predicted in Figure 2, which created the forecast based on data from July 2015 to mid-November 2016, showing the importance of being able to select the data you use. The orange dotted line exhibits the opposite behavior.
Sometimes you know upfront the timeframe on which you want to base your trends. If so, instead of zooming in on a larger data set, an effective trending tool will allow you to select a custom range of dates. This enables you to forecast as far into the future as necessary based on any relevant data you have amassed.
If you’re involved in IT infrastructure strategic or capacity planning, you need a tool you can use to help you create forecasts quickly and easily. Looking into the future helps you plan effectively and ensure optimal user experiences. Make sure that whatever tool you use allows you to select data from the relevant period, whether it’s two years or two months.