The above table shows the current IO500 list as of SC17. By default the merely shows the top-level summary information.
To show more data, such as the individual benchmark scores, use the “Add column” drop-down box above. There are shortcuts, such as ior.*, to add all of the individual benchmarks within a group. The three groups are the bandwidth scores, as measured with ior, the metadata scores, as measured with mdtest, and the find scores (currently just one), as measured with the find utility.
To add a custom (i.e. derived) column, merely enter an equation in the Equation field above.
For equations you may use variables, i.e., use the group name of the metrics you want to use and replace space in the variable name with underscore, for example:
If the metrics name is unambiguous, you may omit the group name, i.e., the following is valid:
The score by which the ranking is computed is the 'score' column. This 'score' column is merely the square root of the product of the 'bw' and 'md' scores. The 'bw' score is the geometric mean of the IOR scores and the 'md' score is the geometric mean of the mdtest scores and the find scores.
Please consider submitting your own result to make it onto the next list! Fetching, compiling, and running the IO500 Benchmarks is surprisingly straightforward. Although tuning can be challenging. Instructions for the former can be found in our github repo. Help with the latter can be found in our various communication channels.