STAC Network I/O Testing Program - SUMMARY PAGE

The STAC Network I/O Testing Program is a series of testing projects, most of which are requested by the STAC Network I/O SIG. These projects generally lock down all the layers of the "stack under test" (SUT) except for one or two, in order to make apples-to-apples comparisons at a given layer. For example, several tests may be run on the same server and OS while varying only the host adapter or network driver. This page is updated as new test results are made available.

Below are links to tests performed using STAC-N tools. (Note: This does not include the STAC-N pilot project, "STAC Network API Study A"). When comparing reports, please note the version of benchmark specs. Different versions cannot necessarily be compared directly.

The table at the bottom of this page indexes the reports that are available from this program to date. You may also be interested in:

• A CSV file containing summary latency statistics for all configurations tested at all message rates. This enables easy analysis within Excel or another tool.

• The latest package of STAC-N test harness software to run your own tests.

• A slide deck covering some optimization findings from testing with RHEL 6.3 and Solarflare.

• A slide deck summarizing the RHEL 6.3 v 5.8 differences with the kernel driver and OpenOnload using Solarflare adapters.

• A slide deck summarizing the RHEL 6.3 v 5.8 differences with the kernel driver and Mellanox adapters.

Each report in the table below corresponds to a unique SUT. Note that a SUT is defined by all aspects of its configuration. That means that any change to the configuration (hardware, software, even a config parameter) defines a new SUT. The columns capture several components of the SUT, but not all. In the event that two SUTs have identical values for all of the components listed, the SUTs are distinguished by the configuration number in the "CONFIG" column at the far right.

* SM = Supermicro