Performance case after normalizing
Normalizing a machine-wide MIPS rating for the number of processors on the machine results in a more representative estimate for comparative transformation performance. It accounts for the number of processors on a box.
Comparing performance
Analyze the following table of the MSU ratings that have been normalized for two different hardware configurations to account for the number of processors on each box:
| Configuration | Number of Processors | Rating (MSUs) before Normalization | Rating (MSUs) after Normalization |
|---|---|---|---|
| Config-02 | two | 28 | 14 |
| Config-01 | one | 13 | 13 |
Conclusion of comparison after normalization
Because the Config-01 configuration has only a single processor, the normalized rating and non-normalized ratings are the same. This similarity in normalized MSU ratings accounts for similarities in execution times. All other things being equal between the two configurations, the two different machines should execute a single transformation, X, in roughly the same time, T.
Parallel execution
Even though the transformation execution times for the two boxes are roughly the same, the Config-02 configuration with two processors, could execute two instances of that same transformation X in parallel. This assumes that the input and output data sets have no interdependence. In this case, the execution time for each transformation is roughly the same.
If both transformations are initiated at the same time, however, they should complete in that same time interval, T. Because both transformations executed in parallel, the amount of transformation work for both is twice as much as it would be for a single execution.
Twice the amount of transformation work in the same amount of time yields double the throughput, which represents a significant performance improvement.