|
Capacity Planning Information
Capacity planning is the key to a well-oiled, smooth-running data center, helping IT organizations to:
- Accurately and efficiently provision new applications
- Reduce, delay and sometimes completely avoid costs
- Justify upgrades with objective analyses
- Optimally configure systems to accommodate forecasted changes in business
- Migrate systems to take advantage of new technology
- Consolidate servers to simplify management and gain economies of scale
- Mitigate service delivery risk by playing out potential scenarios in advance, and then preparing for them
Capacity Planning Techniques
A variety of capacity planning techniques are available, each useful in its own right. TeamQuest provides tools for each of these techniques:
Trending |
Trending uses simple extrapolation of resource utilization over time. The advantage of this technique is that it does not require sophisticated tools. A spreadsheet will do, but it can be difficult to take into account non-linear behavior buried deep within a multi-tiered system. |
Linear Trend
Analysis |
Linear trend analysis looks at historical data and projects a linear trend line into the future, applying upper and lower confidence intervals as well as a threshold at which resources will become inadequate. This capacity planning method is a quick sanity check and identifies over-utilized resources, but it does not allow for experimentation with different configurations. |
Short Term
Analysis |
Short term analysis identifies potential problems early by generating exception events (alarms) based on historical averages (rather than a fixed threshold) tied to a particular point in a periodic time interval. For example, if a particular resource averages 60% busy on Tuesdays and 50% busy on Thursdays, an exception event will only be generated when utilization crosses those thresholds on the respective days. |
Simulation
Modeling |
Simulation, a much more sophisticated capacity planning technique than trending, actually simulates the queuing events that occur during execution. The downside of simulation modeling is that it can be very time-consuming to build and run the models. |
Analytic
Modeling |
Analytic modeling uses mathematics to calculate how a queuing network will perform. To use analytic modeling, a baseline set of measurements is taken on a system and a model is then built based on a description of the system. Results from the model are compared with the baseline, and when they match, the model is considered calibrated. From there, hypothetical changes can be made to system configuration or business workloads, and the model will predict how the changes will affect performance. Done properly, analytic modeling is a fast and accurate capacity planning technique. |
Analytic Modeling and Load Testing
Especially when rolling out new applications, it can be useful to use a load-testing tool in conjunction with an analytic modeling tool.
A load-testing tool can be used to simulate transactions coming from end users, or applications can be benchmarked on test systems to be certain they can support required service levels. However, time and cost considerations often make it prohibitive to conduct such tests on the actual hardware using production-level workloads. Full-sized machines and software are too costly to dedicate for testing purposes, and finding the optimal configuration through empirical testing can take a long time.
Instead, smaller but representative loads can be applied to a scaled-down set of test servers while performance analysis software takes a baseline reading of performance. Analytic modeling can then be used to rapidly predict how various configurations will perform under a production-level workload, all without the need to purchase the actual configurations under consideration. Analytic modeling is a cost-effective, fast and accurate alternative to brute-force testing.
|
Request online demo
