Enterprise Drives Success with Business Metrics, Data, and Automated Forecasting

Enterprise Holdings (EHI) is the parent company of Enterprise Rent-A-Car, Alamo Rent A Car, National Car Rental, and Enterprise CarShare. With annual revenues of $16.4 billion and more than 78,000 employees, EHI and its affiliates own and operate almost 1.4 million cars and trucks. By revenue, employees, and fleet, it is the largest car rental service provider in the world.

In the past, performance and capacity forecasting and modeling at EHI was resource-intensive and error-prone, requiring people to enter large volumes of data into Microsoft Excel or Access manually. This process was repeated quarterly and annually, costing the company dozens of resources and countless hours to collect data, guesstimate growth, and present a forecast.

“One common error committed at that time was to create a forecast using linear trending for CPU usage,” said Performance Engineer Clyde Sconce. “If you do it that way, in our experience, you will be mostly wrong. As too much time was being spent on forecasting with largely poor results, we decided to implement Vityl Capacity Management.”

Forecast Inputs

EHI uses PostgreSQL database information such as server configuration (current and historical), resources consumed (CPU, memory, storage), and business transactions (via user agents). “Our AIX systems also provide plenty of useful metrics such as RPERF, which is a rating for horsepower consumption that helps you to see whether you need to add or remove CPUs,” Sconce said. “These measures normalize capacity across different models and/or platforms, as well as provide continuity when servers are upgraded or migrated so that historical data is not lost.” Hardware configurations planned for the future also need to be incorporated into forecasts.

EHI also uses overrides for several purposes. In some cases growth rates may be inaccurate if historical data is incomplete. For example, a brand new server may show up as having 300% growth. Sconce can override numbers like that in the forecasts, correcting them to a more realistic number. EHI watches for baseline jumps, such as shifts in resource consumption without changes in growth rates. For example, the data may tell an inaccurate story when two servers are merged into one. In that case, the workload has doubled but the growth rate has not changed.

The Key is Business Alignment...

Of course, forecasting can’t be an IT-only activity. Sconce advises it’s essential to take current as well as historical business transactions into account. Business data helps represent how much work is actually consuming resources. A big metric in the world of EHI, for instance, is cars rented per hour.

“The business lives and breathes on that number, so whatever IT metrics we use internally, we always translate into cars per hour when communicating with management. This makes it necessary to correlate business transactions to resources consumed in order to estimate the cost to the organization.”

To ensure accuracy, EHI communicates with the various lines of business to get an estimate of anticipated growth. Sconce emphasizes that this has to be done using metrics that are familiar to the business.

“Taking the metric of dollar cost per server coming from our Accounting Department database, we are able to estimate the actual dollar cost per business application and per transaction,” explained Lead Performance Engineer Gary Savage. “This is invaluable when it comes to translating IT metrics into the language of the business and in seeing how well we are doing in terms of cost.”

IT can group and associate forecasts based on specific groups of servers in various regions or for different functions. The team also harnesses third party vendor tools to collect Java environment statistics as well as transactional data (such as URL hit counts, which offer a good head count).

Savage says they use event tags on their data, allowing them to label historical or even future events that may have significantly altered the forecasting pattern.

“For example, if management asks us about something that happened two years back, we can find it easily by referring to its event tag,” Savage said.

Predictive Analytic Accuracy

Savage says one key to accurate predictive analysis is how the information is processed.

“An accurate forecast must employ a sophisticated analytical tool that can do things like cyclical trending, anomaly removal, baseline shifts, hardware changes, cost correlations and flexible report groupings (by theme, solution, location, etc.). Just throwing all of the data and inputs into a blender won’t work very well. Vityl provides and automates all of these functions.”

Savage and Sconce say there are many important lessons they have learned from their Vityl-based forecasting. “The values we rely on the most, for example, are peak hourly averages at the server level,” Savage and Sconce said. “We have also found it useful to have exception reports generated to flag servers with missing data or anomalies that need to be investigated. Further, we found it best to focus on organic consumption (raw CPU, memory, storage) and conduct overrides for editing of anomalous data, annual growth rate, and baseline jumps.”

For organic, cyclical growth, EHI calculates annual growth (monthly compounded) then applies a cyclical pattern based on monthly usage. “Monthly cyclical growth lets us be more accurate and make more timely capital expenditures when compared to linear (annual) projections. For example, a linear projection may let us know to make a purchase in June. But by using organic cyclic growth, we were able to delay the purchase until December when it would be needed for a seasonal peak,” they said.

Reality Check

There is no “easy button” when it comes to this type of sophisticated predictive analysis, according to Savage. He says it requires implementing a robust forecast reporting process, which takes commitment and many hours of hard work. However, once completed, the benefits are significant.

“We dramatically reduced our staff resource time commitments,” Savage said. “We were able to automate the forecasting process and implement daily/weekly reporting; we developed a standardized forecasting strategy and were able to conduct historical forecast tracking to identify areas of improvement. We were, therefore, able to provide superior forecasting accuracy. We believe that each targeted objective was successful and Vityl clearly demonstrated the ability to achieve or surpass our requirements.”