Search

The Coupled Model Intercomparison Project (CMIP) is one of the biggest international efforts aimed at better understanding the past, present, and future of climate changes in a multi-model context. A total of 21 model intercomparison projects (MIPs) were endorsed in its sixth phase (CMIP6), which included 190 different experiments that were used to simulate 40 000 years and produced around 40 PB of data in total. This paper presents the main findings obtained from the CPMIP (the Computational Performance Model Intercomparison Project), a collection of a common set of metrics, specifically designed for assessing climate model performance. These metrics were exclusively collected from the production runs of experiments used in CMIP6 and primarily from institutions within the IS-ENES3 consortium. The document presents the full set of CPMIP metrics per institution and experiment, including a detailed analysis and discussion of each of the measurements. During the analysis, we found a positive correlation between the core hours needed, the complexity of the models, and the resolution used. Likewise, we show that between 5 %–15 % of the execution cost is spent in the coupling between independent components, and it only gets worse by increasing the number of resources. From the data, it is clear that queue times have a great impact on the actual speed achieved and have a huge variability across different institutions, ranging from none to up to 78 % execution overhead. Furthermore, our evaluation shows that the estimated carbon footprint of running such big simulations within the IS-ENES3 consortium is 1692 t of CO 2 equivalent. As a result of the collection, we contribute to the creation of a comprehensive database for future community reference, establishing a benchmark for evaluation and facilitating the multi-model, multi-platform comparisons crucial for understanding climate modelling performance. Given the diverse range of applications, configurations, and hardware utilised, further work is required for the standardisation and formulation of general rules. The paper concludes with recommendations for future exercises aimed at addressing the encountered challenges which will facilitate more collections of a similar nature. [ABSTRACT FROM AUTHOR]

The Coupled Model Intercomparison Project (CMIP) is one of the biggest international efforts to better understand past, present and future climate changes in a multi-model context. A total of 21 Model Intercomparison Projects (MIPs) were endorsed in its 6th phase (CMIP6), which included 190 different experiments that were used to simulate 40000 years and produced around 40 PB of data in total. This paper shows the main results obtained from the collection of performance metrics done for CMIP6 (CPMIP). The document provides the list of partners involved, the CPMIP metrics per institution/model, and the approach used for the collection and the coordination behind this process. Furthermore, a section has been included to analyze the results and prove the usefulness of the metrics for the community. Moreover, we describe the main difficulties faced during the collection and propose recommendations for future exercises. [ABSTRACT FROM AUTHOR]

Mixed-precision approaches can provide substantial speed-ups for both computing- and memory-bound codes with little effort. Most scientific codes have overengineered the numerical precision, leading to a situation in which models are using more resources than required without knowing where they are required and where they are not. Consequently, it is possible to improve computational performance by establishing a more appropriate choice of precision. The only input that is needed is a method to determine which real variables can be represented with fewer bits without affecting the accuracy of the results. This paper presents a novel method that enables modern and legacy codes to benefit from a reduction of the precision of certain variables without sacrificing accuracy. It consists of a simple idea: we reduce the precision of a group of variables and measure how it affects the outputs. Then we can evaluate the level of precision that they truly need. Modifying and recompiling the code for each case that has to be evaluated would require a prohibitive amount of effort. Instead, the method presented in this paper relies on the use of a tool called a reduced-precision emulator (RPE) that can significantly streamline the process. Using the RPE and a list of parameters containing the precisions that will be used for each real variable in the code, it is possible within a single binary to emulate the effect on the outputs of a specific choice of precision. When we are able to emulate the effects of reduced precision, we can proceed with the design of the tests that will give us knowledge of the sensitivity of the model variables regarding their numerical precision. The number of possible combinations is prohibitively large and therefore impossible to explore. The alternative of performing a screening of the variables individually can provide certain insight about the required precision of variables, but, on the other hand, other complex interactions that involve several variables may remain hidden. Instead, we use a divide-and-conquer algorithm that identifies the parts that require high precision and establishes a set of variables that can handle reduced precision. This method has been tested using two state-of-the-art ocean models, the Nucleus for European Modelling of the Ocean (NEMO) and the Regional Ocean Modeling System (ROMS), with very promising results. Obtaining this information is crucial to build an actual mixed-precision version of the code in the next phase that will bring the promised performance benefits. [ABSTRACT FROM AUTHOR]

Mixed-precision approaches can provide substantial speed-ups for both computing- and memory-bound codes requiring little effort. Most scientific codes have overengineered the numerical precision leading to a situation where models are using more resources than required without having a clue about where these resources are unnecessary and where are really needed. Consequently, there is the possibility to obtain performance benefits from using a more appropriate choice of precision and the only thing that is needed is a method to determine which real variables can be represented with fewer bits without affecting the accuracy of the results. This paper presents a novel method to enable modern and legacy codes to benefit from a reduction of precision without sacrificing accuracy. It consists in a simple idea: if we can measure how reducing the precision of a group of variables affects the outputs, we can evaluate the level of precision this group of variables need. Modifying and recompiling the code for each case that has to be evaluated would require an amount of effort that makes this task prohibitive. Instead, the method presented in this paper relies on the use of a tool called Reduced Precision Emulator (RPE) that can significantly streamline the process. Using the RPE and a list of parameters containing the precisions that will be used for each real variable in the code, it is possible within a single binary to emulate the effect on the outputs of a specific choice of precision. Once we have the potential of emulating the effects of reduced precision, we can proceed with the design of the tests required to obtain knowledge about all the variables in the model. The number of possible combinations is prohibitively large and impossible to explore. The alternative of performing a screening of the variables individually can give certain insight about the precision needed by the variables, but on the other hand some more complex interactions that involve several variables may remain hidden. Instead, we use a divide-and-conquer algorithm that identifies the parts that cannot handle reduced precision and builds a set of variables that can. The method has been put to proof using two state-of-the-art ocean models, NEMO and ROMS, with very promising results. Obtaining this information is crucial to build afterwards an actual mixed precision version of the code that will bring the promised performance benefits. [ABSTRACT FROM AUTHOR]

Earth system models have considerably increased their spatial resolution to solve more complex problems and achieve more realistic solutions. However, this generates an enormous amount of model data which requires proper management. Some Earth system models use inefficient sequential Input/Output (I/O) schemes that do not scale well when many parallel resources are used. In order to address this issue, the most commonly adopted approach is to use scalable parallel I/O solutions that offer both computational performance and efficiency. In this paper we analyse the I/O process of the European Centre for Medium-Range Weather Forecasts (ECMWF) operational Integrated Forecasting System (IFS) CY43R3. IFS can use two different output schemes: a parallel I/O server developed by MeteoFrance used operationally, and an obsolescent sequential I/O scheme. The latter is the only scheme that is being exposed by the OpenIFS variant of IFS. "Downstream" Earth system models that have adopted older versions of an IFS derivative as a component – such as the EC-Earth 3 climate model – also face a bottleneck due to the limited I/O capabilities and performance of the sequential output scheme. Moreover it is often desirable to produce gridpoint-space Network Common Data Format (NetCDF) files instead of the IFS native spectral and gridpoint output fields in General Regularly-distributed Information in Binary form (GRIB) format, which requires the development of model-specific post-processing tools. We present the integration of the XML Input/Output Server (XIOS) 2.0 into IFS CY43R3. XIOS is an asynchronous Message Passing Interface (MPI) I/O server that offers features especially targeted at climate models: NetCDF format output files, inline diagnostics, regridding, and when properly configured, the capability to produce CMOR-compliant data. We therefore expect our work to reduce the computational cost of data-intensive (high-resolution) climate runs, thereby shortening the critical path of EC-Earth 4 experiments. The performance evaluation suggests that the use of XIOS 2.0 in IFS CY43R3 to output data achieves an adequate performance as well outperforming the sequential I/O scheme. Furthermore, when we also take into account the post-processing task, needed to convert GRIB files to NetCDF files, and also transform IFS spectral output fields to gridpoint space, our integration not only surpasses the sequential output scheme but also the IFS I/O server. [ABSTRACT FROM AUTHOR]