COMSOL 6.4 - Joule Heating of a Microactuator — Distributed Parameter Version

Joule Heating of a Microactuator — Distributed Parameter Version

The purpose of this example is to demonstrate how to access the cluster computing functionality in COMSOL from the COMSOL Desktop and use it to submit a batch job to a cluster through a job scheduler. The model takes advantage of the distributed parameter functionality in COMSOL. The model also demonstrates how you can measure the speedup of COMSOL on your cluster. The speedup is defined as the quotient between the total runtime using only one physical node and one core of the cluster and the runtime using several physical nodes and all cores of each physical node of the cluster. For detailed information about the model, see Joule Heating of a Microactuator.

This example model requires a Floating Network License.

COMSOL_Multiphysics/Cluster_and_Batch_Tutorials/thermal_actuator_jh_distributed

Modeling Instructions

Application Libraries

From the menu, choose .

In the window, select > > in the tree.

Click

Study 1

Click the

button in the toolbar.

In the dialog, in the tree, select the checkbox for the node > . With this setting active, is available from the Study node’s context menu.

Click .

Parametric Sweep

In the toolbar, click

In the window for , locate the section.

Click

In the table, enter the following settings:


Parameter name	Parameter value list	Parameter unit
DV (Applied voltage)	range(5,0.5,15)	V

Click to expand the section. Select the checkbox. This setting prepares the model for running a distributed parametric sweep. If you save the model and run it in distributed mode, the compute nodes will work independently in parallel on subsets of the parameters.

Cluster Computing

In the toolbar, click

and choose .

In the window for , locate the section.

Find the subsection. From the list, choose . With this setting, settings specific to the model will be used instead of preference settings.

When applicable, specify the you want to use. If you are testing cluster computing, set the number of nodes to 2 or less first to make sure everything works.

Choose from the following options:

: Use the General scheduler if you intend to submit a job to a job scheduler that you have configured COMSOL to run on

® : Use the Microsoft® HPC Pack scheduler if you intend to submit a job to a Windows HPC Server using the Windows HPC Cluster Manager

: Use the OGS/GE scheduler if you intend to submit a job to a Grid Engine scheduler

: Use the SLURM scheduler if you intend to submit a job to a SLURM scheduler

: Use the PBS scheduler if you intend to submit a job to a PBS scheduler

: Use this setting when you have configured COMSOL to run on job scheduler but only intend to run on a single node of the cluster

In the text field, specify the directory where to store the model.

Make sure you configure the batch directories and COMSOL installation directories correctly. It is good practice to save these values as defaults once you have good settings.

Click

External Process 1

In the toolbar, click

In the section, click to open the file containing the model generated by the batch job associated with this external process in a new COMSOL session.

In the section, the log shows the total solution time. COMSOL automatically takes advantage of all cores; to measure the speedup, set the number of cores to 1 and run a new job.

In the window, under > click .

In the window for , locate the section.

From the list, choose .

Select the checkbox.

Locate the section. In the text field, type batchmodel_np_1.

Next, change the filename to create a new node for the run.

Locate the section. In the text field, enter a name of your choice.

Cluster Computing 1

If you are not using the cluster type , make sure to set the to 1.

In the window, right-click and choose .

Batch 1

In the window, right-click and choose .

External Process

Go to the window.

When the process has finished, compare the total time in the log for the new external process with the previous value. The speedup is equal to the previous value divided by the new value. The speedup depends on the mesh size. To improve the numbers, try refining the mesh. If you are not using the cluster type Not distributed, the speedup also depends on the number of parameter tuples in the Parametric Sweep node. You can also try to increase the number of parameter tuples to improve the numbers.