Each night we run an extensive series of tests on a set of (platform,configuration) pairs. These results can be found on the summary page.
At one point we were looking at a different set of scripts whose functionality overlapped with these tests. The results from one set (now stale) may be seen at a new testing script; this ran some but not all of the tests currently run by the nightly tests but in addition ran several performance tests. Another system, since discontinued, provided web access to all of the log files (including config.log in each directory in which a configure was run), and provide a way to merge results from different platforms. We are still looking for the perfect test system for parallel programs.
Nightly Test Runs:
The tests run from a preset list of configurations present in /home/mpich2/nightly/cron/test.lst on bblogin; we currently have about 52 configurations. There are two rounds of tests each night, one at 9pm and the other at 3am. In each round, the cron job checks how many machines are free and runs that many configurations (each configuration is run on a single machine). If all the machines are free for both the rounds, we can have 48 configurations done. The next night, we start with the next configuration in the list. Another cron job runs at 8am that creates an HTML page out of the last run of each configuration, i.e., all configurations that ran that night + the last available run for each of the other configurations. Finally, a third cron job runs at 8:30am on blundetto that fetches the HTML page from bblogin and places it at http://www.mcs.anl.gov/research/projects/mpich2/nightly/new/latest. There is a backup script running on schwinn that does the third cron job as well at 9:00am, as a backup in case blundetto goes down.
If you need to manually run the nightly tests, you can just execute /home/mpich2/nightly/cron/overall_run.sh.
- Special Tests - This shows the results of special tests, such as configuring, building, and testing with memory tracing (
--enable-g=mem), logging, or shared libraries, on one system (usually Linux). See Running the Special Tests for information on running or modifying these tests. It is a good idea to add tests to this file in response to user bug reports to ensure that fixed bugs stay fixed.
- Tests of Configure Options - This shows the results of configuring, building, and installing with a variety of configure options, on one system (usually Linux). See Running the configure tests for information on running or modifying these tests. It is a good idea to add tests to this file in response to user bug reports to ensure that fixed bugs stay fixed.
We are currently missing the following test result pages from the old website:
- Results from the MPICH test suite - This shows more diverse tests but run against only the MPICH test suite. As an added benefit, these results package up all relevant generated files, include config.log, from each configured subdirectory.
Compilers Tested Regularly: GNU, Intel, PGI, Absoft
(SUN compilers, as well as other vendor compilers, such as the IBM XL compilers, are not included in our nightly tests but are tested on an occasional basis, depending mostly on the availability of a test platform.)