Usability Trade-off Matrix

This table summarizes experiment cost and duration across various cloud platforms, environments, and hardware types. The data is grouped by common environments to provide a high-level summary. You can click on any group row to expand it and view the individual experiments.

Experiments were run on cloud providers aws, azure, and google using environment types cpu and gpu. The test sizes included 4, 8, 16, 32, 64, 128, and 256 nodes/instances.

For each test, 5 iterations were run, and the Duration column is the sum of the time taken for each iteration. While standard deviation is not explicitly shown, it is an important factor for result stability and is reflected in the final Cost calculation. Note that if you want to compare the total experiment cost, you need to select groups with all sizes (N=4) using the dropdown above.