Research Computing @Cambridge

• Cambridge Service for Data-Driven Discovery (CSD3)
  • Wilkes-2
    360 NVidia P100 GPUs
  • PETA4
    24576 cores Intel Skylake +
    342 nodes Intel Xeon Phi (Knights Landing)
  • Openstack private cloud infrastructure

Research Software Engineering @Cambridge

• Partner with researchers at Cambridge and other Universities
• Help researchers make better use of our compute resources
• Spread information about best practices
  "software carpentry" "reproducible research"

What do we want to test with Continuous Integration?

Unit testing

• Test every function
• Test for all possible values
• C++ unit testing frameworks:
• Catch
• Google mock test
• Boost unit test

All tests passed (36 assertions in 3 test cases)

Code coverage

                    
#Codecov
  - compiler: gcc
env:
  - SPECIAL=codecoverage
addons:
  apt:
    sources: ['ubuntu-toolchain-r-test']
    packages: ['g++-5', 'ruby']
  before_script:
    - wget lcov_1.11.orig.tar.gz
    - tar xf lcov_1.11.orig.tar.gz
    - sudo make -C lcov-1.11/ install
  after_success:
    - rm -rf ./* && cmake .. -DENABLE_COVERAGE=On
    - make clean && make
    - ./felib
    - ./felibtest
    - make gcov
    - make lcov
    - bash <(curl -s https://codecov.io/bash) -X gcov
                    
                  

Static analysis

TravisCI: Build matrix for different compilers

Travis allows for Caching repos and directories

Installing multiple versions of libraries is complicated!

Docker containers

• Automates the deployment of applications by packaging an application with all of its dependencies into software containers.
• Docker containers wrap up a piece of software in a complete filesystem that contains everything it needs to run: code, runtime, system tools, system libraries – anything you can install on a server.
• This guarantees that it will always run the same - Research reproducibility

CircleCI: Custom Docker containers

                      
version: 2
jobs:
  build:
    working_directory: /home/cbgeo/research/
    docker:
      - image: quay.io/cbgeo/mpm3d
    steps:
      - checkout
      # GCC 6.3
      - run:
          name: GCC-6 build
          command: |
            mkdir -p build
            [ "$(ls -A build)" ] && rm -rf build/*
            cd build
            cmake -DCMAKE_BUILD_TYPE=Release \
                  -DCMAKE_CXX_COMPILER=g++ ..
            make -j8 
            ctest -VV
                      
                    

Custom docker images

Quay.io / GitLab / Docker hub container registry

Challenges of running Docker on HPC environments

• docker daemon runs as root
• users inside a docker container assume root level privilege
• makes mounting filesystems in a secure way impossible

Shifter "Containers for HPC"

• Project led by NERSC and Cray
• Complex infrastructure required

Singularity containers

• Compatible with (many) docker images
• Installed as a single (setgid) executable
• uid same inside and outside container
• Filesystems can be used without problems

Jenkins + CI on HPCS

                    
singularity run \
    --bind $(pwd):/lammps \
    --bind /home/jenkins/lammps-run:lammps-run \
    --bind /home/jenkins/lammps-dem-validation:lammps-dem-validation \
    /home/jenkins/centos.img \
    /lammps-run/compile.sh /lammps/src /lammps-dem-validation
                      
                    

Running Singularity images

External slave node connected via JNLP

Challenges of running CI on HPC

• Job schedulers