The following information is provided for each benchmark:

• name: the name of the benchmark
• layers: this covers the storage API that is actually benchmarked, e.g. POSIX
• features: Describes roughly the offered features that are supported (potentially via options) by the benchmark
  • data: This is a data centric benchmark (read, write)
  • metadata: This benchmark measures metadata performance (open, close, stat, readdir, …)
  • programmable: The details of the access pattern can be configured, e.g., execution order.
  • tunable: Offers fine grained access to the underlying API supporting many tunables (e.g., hints)
  • analyzable: The benchmark suite provides at least rudimentary support to visualize and analyze the results and compare several runs.
• parallel: This indicates if and how the benchmark is parallelized:
  • Sequential, i.e., a single process that runs the benchmark
  • OpenMP: Parallelized via OpenMP
  • MPI: MPI parallel version
• type: the type of the access pattern.
  • synthetic: completely artificial access pattern
  • kernel: this is an I/O kernel of an existing application, meaning it only contains the I/O parts and runs them as fast as possible
  • mini-app: this is a stripped down version of an application consisting of I/O but also compute phases
  • application: this is a full featured scientific application that does make sense
• license: The license behind the benchmark
• webpage: The web page of the benchmark

This page contains a list of popular benchmarks; by clicking a particular characteristics, the list is filtered. Multiple characteristics can be selected.