LES of Hurricane Boundary Layer

In this tutorial, we demonstrate how to simulate a simplified hurricane boundary layer (HBL) using Large Eddy Simulation (LES) on the Inductiva platform. This example was retrieved from the official CM1 repository 1.

The hurricane boundary layer is the lowest portion of the atmosphere in a tropical cyclone where turbulence, surface friction, and radial inflow are dominant. This LES setup simulates the internal dynamics of the HBL without full coupling to large-scale weather systems.

We will also demonstrate Inductiva’s ability to efficiently scale this use case, starting with a cloud machine equivalent to a typical laptop and then scaling up to more powerful machines.

Prerequisites

Download the required files here and place them in a folder called cm1-les-hurr-bound-layer.

Run the Simulation

Below is the code required to run the hurricane boundary layer use case with the Inductiva API.

The simulation will then be sent to a c2d-highcpu-16 virtual machine from Google Cloud, equipped with 16 vCPUs and 32 GB of RAM. This machine is equivalent to a standard working laptop.

"""LES of Hurricane Bound Layer."""
import inductiva

# Allocate cloud machine on Google Cloud Platform
cloud_machine = inductiva.resources.MachineGroup(
    provider="GCP",
    machine_type="c2d-highcpu-16",
    spot=True,
)

# Set simulation input directory
input_dir = inductiva.utils.download_from_url(
    "https://storage.googleapis.com/inductiva-api-demo-files/"
    "cm1-les-hurr-bound-layer.zip",
    unzip=True,
)

# Initialize the Simulator
cm1 = inductiva.simulators.CM1( \
    version="21.1")

# Run simulation with config files in the input directory
task = cm1.run(
    input_dir="/path/to/cm1-les-hurr-bound-layer",
    sim_config_filename="namelist.input",
    on=cloud_machine,
)

# Wait for the simulation to finish and download the results
task.wait()
cloud_machine.terminate()

task.download_outputs()

task.print_summary()

Copy and paste it into a file named run.py in your working directory and execute it by running:

python run.py

When the simulation is complete, we terminate the machine, download the results and print a summary of the simulation as shown below.

Task status: Success

Timeline:
    Waiting for Input         at 09/07, 14:08:46      0.676 s
    In Queue                  at 09/07, 14:08:47      58.479 s
    Preparing to Compute      at 09/07, 14:09:45      4.285 s
    In Progress               at 09/07, 14:09:49      5772.233 s
        └> 5771.837 s      /opt/openmpi/4.1.6/bin/mpirun --use-hwthread-cpus cm1.exe namelist.input
    Finalizing                at 09/07, 15:46:02      6.8 s
    Success                   at 09/07, 15:46:08

Data:
    Size of zipped output:    617.07 MB
    Size of unzipped output:  1.21 GB
    Number of output files:   776

Total estimated cost (US$): 0.16 US$
    Estimated computation cost (US$): 0.15 US$
    Task orchestration fee (US$): 0.010 US$

Note: A per-run orchestration fee (0.010 US$) applies to tasks run from 01 Dec 2025, in addition to the computation costs.
Learn more about costs at: https://inductiva.ai/guides/basics/how-much-does-it-cost

As you can see in the "In Progress" line (the part of the timeline that represents the actual execution of the simulation), the core computation time of this simulation was approximately 1 hour and 36 minutes (5772 seconds).

Scaling Up

One of the strengths of running LES simulations on the Inductiva platform is the ability to scale your workload across a wide range of high-performance cloud machines. Whether you need faster results or better cost-efficiency, scaling up is as simple as adjusting the machine_type parameter when allocating your cloud machine.

We tested the same simulation across several cloud machines with increasing vCPU counts and next-generation architecture. The results show a clear trend: more powerful machines can dramatically reduce execution time while maintaining reasonable costs.

Generational improvements matter. Upgrading from c2d-highcpu-16 (a typical entry-level machine) to c4-highcpu-16 (a newer generation with the same number of vCPUs) improved runtime by 45%.

Increasing the number of vCPUs to 112 on c2d-highcpu-112 slashed the execution time by over 5×, with only a slight increase in cost—offering. It's one of the most cost-effective high-performance options.

For maximum performance, c4-highcpu-192 brought the simulation time down to just 11 minutes, achieving over 8× speedup compared to the baseline. It delivers exceptional speed for time-critical workloads.

These results demonstrate that Inductiva not only supports rapid scaling but also gives users flexibility to optimize for speed, cost, or a balance of both, depending on the demands of the task. 🚀

If you want to benchmark your own workload in a single script please follow this tutorial.

References

1 Simple hurricane boundary layer (HBL)