We have gitlab-runner running on the machine 'mozart'.
When a test campaign is needed, this program runs the script
.gitlab-ci.yml found in the top directory of the source code.
Activation of a campaign is manual. It's not possible to run
a campaign for each commit because it takes too much time to
run.

The script .gitlab-ci.yml in turn runs the python program
cmake_targets/autotests/v2/main.py with a list of tests to run. Several
files in cmake_targets/autotests/v2/ are used when tests are run. Logs
are stored in cmake_targets/autotests/log/.

The tests are defined in the files cmake_targets/autotests/README.txt
and cmake_targets/autotests/test_case_list.xml.

This version (v2) is a rewrite of the previous one, to speed it up
and remove many bugs. It does not fully use
cmake_targets/autotests/test_case_list.xml.

There are several kinds of tests:
- compilation
- oaisim / phy simulators (ulsim, dlsim)
- eNB with COTS UE

For each test, we have a set of machines to use. The python program
connects with ssh to them and runs shell scripts (found in
cmake_targets/autotests/v2/actions/), collecting terminal outputs
and log files (if any) for later analysis.

As of today the eNB tests use a third party EPC provided by Nokia.
It's possible to adapt the scripts to use openair-cn instead.

The eNB tests that are done are:
- monolithic FDD mode with 1 COTS UE and B210:
  - 5, 10 and 20 MHz
  - run iperf uplink and downlink, tcp and udp, in turn, for 10s
    (default running time of iperf)
- monolithic TDD mode with 1 COTS UE and B210:
   - same tests as for FDD
- split eNB, FDD with if4p5 split and 1 COTS UE and B210:
   - same tests as for FDD

The COTS UE is in a Faraday cage.

A full campaign takes less than one hour. (But the number of tests
done is limited. Also, the previous version ran a test several times.)

The results are stored in cmake_targets/autotests/log/ in the gitlab-runner
directory.

What is missing:
- handle more cases, especially multiple UEs, and other RF equipment
  (X310, bladeRF, limeSDR, ...), run longer, maybe in parallel if
  more equipment is available (several UEs, several RF, ...).
- automatic analysis of the results and production of a summary
  (webpage?) easy to check. The previous version had one, I didn't have
  time to finalize this part. Today, I manually inspect the output
  files and decide if the test is passed or not.

All this is open to improvments, modifications, suggestions.

Cedric Roux - 2017-11-15.