HPC platform modelling Sample Clauses

HPC platform modelling. To estimate the running cost for a given workload, we need a model for its computational environment. This model permits to detail consumption on numerous aspects of a workload execution: • Computing node allocation time • Required storage on each tier • Remote visualization strategy • Data transfer volumes • Cluster service • Support plan Each of these aspects, will correspond to a particular resource type, and a price for the cost estimation. Thus, to describe a computational environment, we introduce the HPC Platform Model in CloudSHaper. Our HPC Platform Model relies on the following core components: • Compute • Storage • Visualization • Service • Data transfers These components are generic enough to encompass the need to describe both on-premises HPC Centres and clusters deployed in a CSP. 1 ▇▇▇▇▇://▇▇▇.▇▇▇ An instance of an HPC Platform Model, to describe an HPC cluster, is called an HPC Cluster Configuration. One can create multiple HPC Cluster Configurations, for each environment he has access to, such as: • An on-premises cluster or an HPC Centre • A cloud cluster hosted on AWS • A cloud cluster hosted on Azure An HPC Cluster Configuration can be accessible in multiple Regions, allowing to describe a model once and providing the capability to describe multiple clusters with multiple pricings. In a CSP, the region directly maps with their notion of Region and Availability Zones (i.e., a geographical location in which the CSP has datacentres, e.g., eu-west-1 for AWS), for an HPC Centre this notion can be used to describe either multiple offerings or to split the pricing per queues for example. Each HPC Cluster Configuration will be used by CloudSHaper as a guideline to breakdown the costs of a given workload and provide a cost estimation to run it on the targeted environment. To allow CloudSHaper to project resource consumptions of a workload on various HPC Cluster Configurations, our HPC Platform Model relies on a series of generic properties for each core component.