Traditionally, a capacity forecast is presented on a monthly basis and has the tendency to represent the capacity trends in a straight predictable line. This paper intends to show the value of measuring system capacity on a weekly basis. It also introduces the use of Process Control Analysis to confirm the forecasting methodology. The information presented has been used and proven in real world operations.
Capacity planning has long been viewed as an art combined with science. The image of an old wizard gazing into the crystal ball to divine the future or using a secret formula that produces a capacity forecast comes to mind. In today’s constantly changing environments, and IT budgets being scrapped to the bone, squeezing every last iota of performance out of a machine is needed and warranted. And now imagine a capacity forecast so reliable and with a high degree of certainty and accuracy where an upgrade can be pin pointed to a specific week of the year, not just the month, and can be used up to twelve months out? Management now has the information it needs to provide “just in time” capacity planning and can control the impact to budget dollars.
The author is going to draw on a methodology that was developed over time and successfully implemented with a high degree of success. The capacity forecast methodology over time has been so accurate that it was able to pinpoint the specific week of the year when an upgrade was needed. The methodology used to achieve these results uses a combination of seasonal forecasting with linear regression. The methodology is documented in the paper, “The Straight Capacity Line.”[i]This methodology has been used for over ten years and has a proven track record. Not only has it been used in the mainframe world, but has been used in a zLinux environment.
[i]CMG 2006, “The Straight Capacity Line”, Linda J Carroll, IBM