Use of several Cloud Computing approaches for climate modelling: performance, costs and opportunities
Cloud Computing is a technological option that offers great possibilities for modelling in geosciences. We have
studied how two different climate models, HadAM3P-HadRM3P and CESM-WACCM, can be adapted in two
different ways to run on Cloud Computing Environments from three different vendors: Amazon, Google and
Microsoft. Also, we have evaluated qualitatively how the use of Cloud Computing can affect the allocation of
resources by funding bodies and issues related to computing security, including scientific reproducibility.
Our first experiments were developed using the well known ClimatePrediction.net (CPDN), that uses BOINC,
over the infrastructure from two cloud providers, namely Microsoft Azure and Amazon Web Services (hereafter
AWS). For this comparison we ran a set of thirteen month climate simulations for CPDN in Azure and AWS using
a range of different virtual machines (VMs) for HadRM3P (50 km resolution over South America CORDEX
region) nested in the global atmosphere-only model HadAM3P. These simulations were run on a single processor
and took between 3 and 5 days to compute depending on the VM type.
The last part of our simulation experiments was running WACCM over different VMS on the Google
Compute Engine (GCE) and make a comparison with the supercomputer (SC) Finisterrae1 from the Centro de
Supercomputacion de Galicia. It was shown that GCE gives better performance than the SC for smaller number of
cores/MPI tasks but the model throughput shows clearly how the SC performance is better after approximately
100 cores (related with network speed and latency differences).
From a cost point of view, Cloud Computing moves researchers from a traditional approach where exper-
iments were limited by the available hardware resources to monetary resources (how many resources can be
afforded). As there is an increasing movement and recommendation for budgeting HPC projects on this technology
(budgets can be calculated in a more realistic way) we could see a shift on the trends over the next years to
consolidate Cloud as the preferred solution.
keywords: Climate prediction, cloud computing, Amazon, Google, Microsoft
Publication: Congress
1624015044600
June 18, 2021
/research/publications/use-of-several-cloud-computing-approaches-for-climate-modelling-performance-costs-and-opportunities
Cloud Computing is a technological option that offers great possibilities for modelling in geosciences. We have
studied how two different climate models, HadAM3P-HadRM3P and CESM-WACCM, can be adapted in two
different ways to run on Cloud Computing Environments from three different vendors: Amazon, Google and
Microsoft. Also, we have evaluated qualitatively how the use of Cloud Computing can affect the allocation of
resources by funding bodies and issues related to computing security, including scientific reproducibility.
Our first experiments were developed using the well known ClimatePrediction.net (CPDN), that uses BOINC,
over the infrastructure from two cloud providers, namely Microsoft Azure and Amazon Web Services (hereafter
AWS). For this comparison we ran a set of thirteen month climate simulations for CPDN in Azure and AWS using
a range of different virtual machines (VMs) for HadRM3P (50 km resolution over South America CORDEX
region) nested in the global atmosphere-only model HadAM3P. These simulations were run on a single processor
and took between 3 and 5 days to compute depending on the VM type.
The last part of our simulation experiments was running WACCM over different VMS on the Google
Compute Engine (GCE) and make a comparison with the supercomputer (SC) Finisterrae1 from the Centro de
Supercomputacion de Galicia. It was shown that GCE gives better performance than the SC for smaller number of
cores/MPI tasks but the model throughput shows clearly how the SC performance is better after approximately
100 cores (related with network speed and latency differences).
From a cost point of view, Cloud Computing moves researchers from a traditional approach where exper-
iments were limited by the available hardware resources to monetary resources (how many resources can be
afforded). As there is an increasing movement and recommendation for budgeting HPC projects on this technology
(budgets can be calculated in a more realistic way) we could see a shift on the trends over the next years to
consolidate Cloud as the preferred solution. - Diego A. Perez Montes, Juan A. Añel Cabanelas, David C. H. Wallom, Alberto Arribas, Peter Uhe, Pablo V. Caderno, and Tomas F. Pena
publications_en