On-Demand Deployment of VMIs for Scientific Application Execution

The execution of scientific applications requires using Virtual Appliances (VAs) which encapsulate the required hardware and software configuration to satisfy the proper execution of the application. The VMRC catalog is employed to index the VMIs with the appropriate metadata that describes their OS and applications.

The scientific jobs to be executed incorporate application requirements (OS requirements, application dependencies, etc.) and are submitted to a Cloud Enactor which uses the job's requirements to query the VMRC catalog for the most appropriate VMIs to execute the applications. The VMIs are stored in OpenNebula's own repository. The reminder sofware not available in the VMIs is deployed using a DevOps tool (i.e. Ansible) at runtime. This procedure enables to deploy an infrastructure to support the execution of scientific applications. This procedure is supported by the Infrastructure Manager (IM) which contacts the VMRC for VMI matchmaking.

On Demand Deployment of VMIs for Scientific Application Execution

Customized Virtual Infrastructures

The GRyCAP provides several on-premises Clouds based on OpenNebula and OpenStack to support the research activities. Different projects require different configurations and a subset of these configuration is available in VMIs (e.g. Ubuntu 14.04, CentOS 6, etc.). Specific software packages are pre-installed in certain VMIs to avoid the delay of installation at runtime.

VMIs are indexed in VMRC. Researchers use the Infrastructure Manager (IM) to deploy their own customized virtual infrastructures. The IM uses the VMRC to know where the most appropriate VMIs are available (in which on-premises Cloud) and to perform the provision and configuration of those computational resources. The VMRC is also employed to catalog AMIs available in AWS (Amazon Web Services), in order to use the IM to perform Cloud bursting.

One of such examples is the case of CodeCloud: A platform to enable execution of programming models on the Clouds a work published in the Journal of Systems and Software, in which is presented a platform that supports the execution of scientific applications covering different programming models (such as Master/Slave, Parallel/MPI, MapReduce and Workflows) on Cloud infrastructures. The VMRC is employed to transform from high-level programming model to the specific VMI which includes the support to execute applications written for that programming model. The figure (at the right) describes the architecture of the CodeCloud platform.