Runtime settings

What you’ll learn

• Node type & count: Choose the right compute resources for your simulation.
• Estimate node count: Use the built-in estimator to determine resource requirements.
• Timeout & core hours: Set time limits and understand core hour consumption.

Runtime settings control the computational resources allocated to a simulation and how long it is allowed to run.

Node type & count

Node type and count are selected in simulation runtime settings:

The node types currently available are:

• Fast start (3 CPU, 10 GB)
  • For small simulations
  • Eliminates startup delay
  • Node count limited to 1
• 1 CPU, 16 GB
  • Default node type
• 2 CPU, 32 GB
• 4 CPU, 64 GB
• 8 CPU, 128 GB
• 16 CPU, 256 GB
• 32 CPU, 512 GB

Node count is an integer, limited by the parallel cores in your plan. For the smallest plan type, which includes 16 parallel CPUs, a maximum of sixteen 1 CPU, 16 GB nodes or one 16 CPU, 256 GB node can be used simultaneously.

Choosing node type & count

Ideally, you should start with the smallest node count and type. Select more or bigger nodes if you need them due to memory limitations.

For multinode simulations, a rule of thumb is to have 50-100k DOFs (degrees of freedom) per 16 GBs of memory. So for 1 million DOFs, 10-20 nodes with 1 CPU, 16 GB would hit the sweet spot.

These are rough estimates — the required memory varies depending on the problem. The direct solver uses more memory than the iterative solver, so for the same problem some more nodes are required by comparison.

Eigenmode analysis

For eigenmode analysis, it’s best to use the direct solver with several 1 CPU, 16 GB nodes. The eigenmode solver is a type of iterative solver, so using the iterative solver with eigenmode analysis applied makes for a very large number of iterations.

Estimate node count

The number of nodes required to run a simulation depends on several factors:

• mesh quality
• selected physics
• analysis type
• solver settings

To gain an estimate on the number of nodes needed for a simulation, use Estimate node count.

How does Estimate node count work?

• The Estimate node count works by running the actual simulation script; except the solving section is replaced by formulation analysis which estimates the required node count.
• The estimation is always run on a single 1 CPU, 16 GB node.
• There is 5-minute timeout for node count estimation and hence the core hour consumption is minimal.
• Node count estimation only corresponds to the main formulation of the physics. If some of the project physics are either pre-solved or post-solved with respect to the main formulation, then those are not taken into account.

When does Estimate node count fail?

• If Full scripting mode is enabled.
• If a single 1 CPU, 16 GB instance has insufficient memory to load the mesh.
• If inserting new code or disabling sections in simulation script causes the simulation run to fail.

Timeout

Timeout determines the maximum time a simulation, sweep or otherwise, is allowed to run. If the simulation is not converged within the timeout period, it will be aborted.

Selecting the timeout duration in Minutes, Hours, or Days.

Maximum used core hours

The maximum used core hours $CH_{max}$ is calculated as

$$\begin{equation*} CH_{max} = \text{Node count} \times \text{number of CPUs in Node type} \times \text{Timeout in hours} \end{equation*}$$

$CH_{max}$ shows the maximum core hours to be consumed if the simulation continues to run for the whole duration set in the timeout period.

When the Node count is more than one, you can choose a bigger Main node type if the mesh is very large. Then the maximum used core hours is calculated as

$$\begin{align*} CH_{max} = & [\text{(Node count - 1)} \times \text{n of CPUs in Node type}] \times \text{Timeout in hours} \\ & + [(\text{1} \times \text{n of CPUs in Main node type})] \times \text{Timeout in hours} \end{align*}$$

Node count = 1	Node count > 1

Note

The Maximum used core hours shown after a simulation is always rounded off to the closest integer.