ESiWACE focuses on preparing Earthâ€system models and tools for the exaâ€scale, and on supporting the weather and climate modelling communities to operate these systems efficiently on a variety of HPC systems. The key target is to perform global cloud and ocean eddy...
ESiWACE focuses on preparing Earthâ€system models and tools for the exaâ€scale, and on supporting the weather and climate modelling communities to operate these systems efficiently on a variety of HPC systems. The key target is to perform global cloud and ocean eddy resolving simulations at 1â€km scale and at a rate of 1 simulated year per calendar day. This will allow breaking down some of the greatest scientific challenges thanks to better predictions of climate trend and much more fidelity in the representation of highâ€impact regional weather and climate events. This is clearly not achievable today and will require exaâ€scale computing and data handling capability.
Initial results from code optimisation by implementing singleâ€precision versions of models, introducing additional levels of parallelism, exploring concurrent hence parallel execution of physics schemes and trialling enhanced programming models show that substantial efficiency gains can be obtained. However these gains alone will not be sufficient to achieve the required runtime targets. ESiWACE is set to implement a longâ€term strategy including incremental as well as potentially disruptive measures to support the weather and climate community to achieve those aims beyond the end of the project.
One key achievement in the reporting period has been the swift inclusion of a new task on very highâ€resolution model simulations—soâ€called demonstrators—to investigate the “computability†of the target resolution and target production rate. This will determine the discrepancy between current and required code performance and will feed into a roadmap proposing the steps to close this gap. Already, a first set of 1â€km resolution runs with both IFS and ICON global atmospheric models as well as a 10â€km coupled oceanâ€atmosphere run with ECâ€Earth have been performed.
In addition to making Earthâ€systemâ€models “exaâ€scale readyâ€, ESiWACE aims at improving their usability especially in view of their exaâ€scale application. The aim is to identify, design and support the endâ€toâ€end workflow for climate modelling and weather forecasting applications in both research and production mode. This will provide the increased human efficiency that will be required to exploit increasingly complex exaâ€scale systems. ESiWACE handbooks, which will constantly evolve throughout the project, support creation of the environment necessary for the communities modelling and system software stacks, Further, we foster the automation of scientific workflows by delivering development and support services to the community for the metaâ€scheduler Cylc. This will also facilitate exploitation of results.
The amount of data that will be generated by highâ€resolution Earthâ€system model runs will be unprecedented and efficient exploitation of data storage will become a major challenge. One of the primary objectives of ESiWACE therefore is to develop tools to enhance and support exploitability of data sets arising from next generation weather and climate models – and to use prototypes of the software system components to handle data from the demonstrator simulations.
First, we addressed requirements and basic (coarseâ€grained) modelling of performance, cost and resilience for weather and climate data storage systems, introducing several alternative scenarios for the architecture of data centres. Already, DKRZ and STFC have started to adopt these architectures. Second, we worked on designs for the architecture of a software â€middleware†layer, which can be implemented between user applications and storage systems aiming at increased performance by local customisation. This work is driven by gathering community requirements as well as the development of an API that will reduce the impact on endâ€users and developers. A third task is the development of a prototype for fineâ€grained simulations of hierarchical storage systems with a focus on tape and mode
A detailed overview of the work performed an be found in the progress report (PR1) on https://www.esiwace.eu/results/misc
A detailed overview of the progress beyond the state of the art and of the expected potential impacts can be found in the progress report (PR1) on https://www.esiwace.eu/results/misc
More info: http://www.esiwace.eu.