HPCBOX’s support for Docker containers was one of the most requested features from our users and we are very excited to bring this unique advantage to our end users. Users can now create, manage, deploy and execute their complex application pipelines by bringing any application they use for their innovation as Docker containers to their HPCBOX cluster.

HPCBOX delivers an innovative desktop-centric workflow enabled platform for users who want to accelerate innovation using cutting edge technologies that can benefit from High Performance Computing Infrastructure. It aims to make supercomputing technology accessible and easier to use for scientists, engineers and other innovators who do not have access to in-house datacenters or skills to create, manage and maintain complex HPC infrastructure. To achieve a high level of simplicity in delivering a productive HPC platform, HPCBOX leverages unique technology that not only allows users to create workflows combining their applications and optimization of underlying infrastructure into the same pipeline, but, also provides integration points to help third-party applications directly hook into its infrastructure optimization capability. Some of the unique capabilities of HPCBOX are:

  • Elastic scale-in and scale-out of HPC clusters
  • Hardware accelerated remote 3D for visualization
  • Rich desktop experience
  • Secure access to resources
  • Individual HPC clusters for every user
  • Intelligent workflow capability
  • Native Docker support

Docker Support

HPCBOX is probably the first HPC Cloud platform to support MPI, OpenGL and CUDA Docker containers on the same platform, letting users create, manage, deploy and execute their entire application pipelines with Docker containers. Unique capabilities offered by HPCBOX for Docker:

  • Support for public and private Docker registries, including Docker Hub and NVIDIA GPU Cloud.
  • Local Docker registry to build your Docker containers directly on the HPCBOX Platform.
  • Support for MPI applications packaged as Docker containers. MPI is over a high speed Infiniband fabric.
  • Support for OpenGL applications packaged as Docker containers. The applications can use hardware accelerated OpenGL rendering for optimum visualization experience.
  • Support for CUDA applications packaged as Docker containers.
  • Seamless creation of complex application workflows combining different types of Docker containers within the same pipeline.
  • Rich desktop experience

CUDA Applications

Applications which use CUDA for computation can be built as Docker containers and pulled on to the HPCBOX cluster from public and private Docker registries.

Multiple instances of these containers can be launched at the same time, in both exclusive and non-exclusive mode. When using exclusive mode, one instance of the Docker container is started on every node eligible to handle CUDA computations. All instances receive an environment variable, HPCBOX_TASK_ID, set to their instance id. This variable can be used for monte-carlo type simulations which run multiple instances of the same application, albeit, with different input files for each instance.

Example running NVIDIA nbody benchmark as a Docker Container

CUDA type Docker container workflow screen

CUDA images available on this demo system.

OpenGL Applications

Visualization applications need hardware accelerated OpenGL capability for optimum user experience. HPCBOX has extended its remote hardware accelerated graphics technology to support OpenGL applications which are packaged as Docker containers. With this innovation, users can pull in and launch any OpenGL application which can use NVIDIA graphics cards. The HPCBOX platform will make sure the NVIDIA graphics card is made visible within the container during execution time. Furthermore, applications like Paraview with the NVIDIA IndeX plugin can be seamlessly run in client server mode within HPCBOX’s rich desktop experience.

Furmark OpenGL benchmark as a Docker container

Paraview with NVIDIA IndeX

MPI Applications

Docker containers with compatible MPI applications can be launched on a high-speed Infiniband fabric on the HPCBOX platform. Users can build MPI applications on their local workstations and pull them on to the HPCBOX platform for running at scale. Alternatively, users can use the HPCBOX platform for creating their MPI applications and push them to other locations using their Docker registries.

Intel MPI benchmark example run from a Docker container

HPCBOX delivers a highly personalized, workflow-enabled rich desktop experience for HPC users. This experience combined with support for Docker containers makes our HPC Cloud platform extremely flexible and one of the most innovative platforms available today.