Computation of performance predictions
Performance predictions typically improve over time due to the following evolutionary stages:
▪Stage 1, estimation: A new system type or model is first introduced. Limited information is available regarding its performance capabilities. It is
possible that a product Announce event can occur before full calibration is completed. So, performance might be estimated based on existing system types and models. For example, developers
can build an estimation by comparing various internal components and processes in the new system to the old. The resulting performance predictions might not be ideal as far as accuracy.
Tip: Monitor the release notes for initial release and upgrades for storage devices. Also monitor the release notes for Storage Modeller itself (see
“Notifications” on page 4).
Watch for changes that might affect capacity and performance results. Typically Stage 1 estimations are conservative. However, unforeseen factors can skew estimations to be overly
optimistic. Watch for gaps that might arise between the estimates of Stage 1 and the more realistic Stages 2 and 3.
|
▪Stage 2, lab results: New system types or models become better understood, as detailed and exhaustive performance measurements are made in lab
environments. Evaluations are made regarding the behavior of the system with different configurations under all manner of environmental conditions. Results of such measurements can be used
to “calibrate” the internal Storage Modeller performance models for the system, which will then yield much more accurate predictions. Most storage system types that Storage Modeller
supports fall into this category.
▪Stage 3, field results: The calibrated internal performance models can be further refined based on observed system behavior in the real world, as
opposed to in lab environments. Such refined calibrated performance models yield the most accurate predictions.
|
Note: Take into account an end-of-life Stage 4, no longer able to purchase. When a product is withdrawn from
marketing, full-calibration activity ceases.
|
The Storage Modeller performance reports include a Performance Model Confidence diagram, which graphically depicts the relative confidence (accuracy) of the performance predictions for
each of the modeled systems in a project. For details, see
“Performance confidence diagram” on page 54.