benchmark Python 3.x

[orbeckst]

At the moment we only run Python 2.7 (benchmarking from release 0.11.0 onwards).

However, since we are fully supporting Python 3 since 0.17.0, we should add benchmarks starting from release 0.17.0 that run under Python 3.6. We still need to run a cython matrix #7 but the results might be interesting because I don't know if there's any performance difference between Python 2 and 3.

Things to do

• make sure that we only run 3.6 benchmarks since 0.17.0: in benchmarking/benchmarks/scripts/cron.sh

   timeout 20h asv run -e -j 4 --config ${ASV_CONFIG} "release-0.17.0..HEAD --merges" --skip-existing || true

• change matrix to 3.6 in ASV_CONFIG="${ASV_RUN_DIR}/asv_c3potato.conf.json"

  "pythons": ["3.6"]

• disable cron job until history is built up
• manually run to build a history from the earlier merge commits
• test that new results appear in https://www.mdanalysis.org/benchmarks/
• test cron job
• update wiki docs (ongoing)
  • benchmark on c3potato
  • automated benchmark

[orbeckst]

I can just add Python 3.6 to the matrix but that does not give me a convenient way to restrict it to specific commits.

Perhaps I'll just let it run and let ASV run as many tests as it can get to run... (however, the initial backlog will probably take almost a week to run).

[orbeckst]

The 3.6 history is currently being run – will take about 5 days.

[orbeckst]

(Continue from MDAnalysis/mdanalysis#2747 (comment))

EDIT: moved to main issue text

[orbeckst]

Re-open until the cron job is running again.

[orbeckst]

I pushed partial results with Python 3.6 to https://www.mdanalysis.org/benchmarks/. Sometimes Python 3.6 is faster, sometimes it's slower. Once we have more data, we should look into which Python 3.6 constructs are problematic. For example, all PBC-related tests are slower in 3.6. For others the picture is mixed.

[orbeckst]

The initial backlog was ran and we now have Py 3.6 benchmarks since ~0.17.0. The cron job should be working.