VMTKLab CFD on the Cloud: User Manual

With VMTKLab and the CFD on the cloud module powered by Rescale you can run computational fluid dynamics simulations on cloud computing HPC resources. Easily set boundary conditions in the GUI and request a number of simulations simultaneously as needed by large-scale study applications.

VMTKLab CFD on the Cloud: User Manual

Once entered in the workflow, user has to choose between: - Dashboard, where she can handle simulations.- New simulation, where she can create a new simulation ( job ) on the cloud.

Click on “Launch a new simulation”.

User has now to load a volume mesh file. The system will check the input file for consistency. Volume mesh model needs to have boundary ids and a valid topology, if boundaries are missing or the topology is invalid an error message will appear. If the file is loaded successfully, user can click on the continue button.

VMTKLab CFD on the Cloud: User Manual

User has now to setup the boundary conditions. Before doing that, she has to provide information about model dimension. For each boundary the mean vessel radius is displayed, allowing the user to choose the unit of measurement.

VMTKLab CFD on the Cloud: User Manual

User has to choose the solver type (at the moment, v1.5.0, multiDGetto solver is the only one available) and provide boundary conditions and simulation parameters. User has to setup at least one inlet providing information about mean flow rate and waveform. Waveform can be steady, pulsatile or user-defined with a txt file.

The input waveform txt file has to be a comma-separated file with time and flow value in mL/min. Output boundary condition waveform will be generated runtime from inlet (computed). Accept boundary conditions button will be enabled once all preconditions have been satisfied.

VMTKLab CFD on the Cloud: User Manual

Click on “Accept Boundary conditions” and then click on “Hardware configuration” on the top right corner of the menu. User has now to select how many cores she’d like to use, setup a name, select if the job will run under a low priority or not and if the job will run with a fast preview run. The Low Priority setting allows you to run your job at a reduced price point, but comes with some restrictions. Overall, the reliability of low priority mode is lower than that for on-demand and selecting this mode could cause a maximum delay in run-time of one time the job's run-time. As a result, the low priority option is not generally recommended for jobs with long run times or subject to upcoming deadlines.

The preview test mode setting allows you to run your job with only 1 cycle and 100 timesteps, with a very low resolution. The advantage is that user will not be charged with the 2€ fixed cost of the simulation. This setting is recommended if the user will like to check the outcome of a simulation before submitting it in a full resolution mode.

User finally clicks on “Create the CFD Job” and the job will be uploaded on the cloud platform.

VMTKLab CFD on the Cloud: User Manual

In the top bar of the window user can see her residual credit. Each job has: -  id, the unique identifier of the job, assigned by the server. -  name, the user-defined name, not unique. -  date started, the time point in which the job has started. -  date ended, the time point in which the job has ended. -  elapsed time, the number of hours to pay for this job. -  cost/hour, the cost for each hour for this job. -  max runnable time, the number of hours you can afford to run this job with your credit. -  status: - ready, user can launch the job by clicking on the action icon. - running, user can stop the job by clicking on the action icon. - stopping, user has to wait for the server. - completed, user can download results by clicking on the action icon. - downloaded, user has already downloaded the results. -   info, if a job is running user can click this icon and see the process output log of the simulation. -   actions, icon to be clicked for launching/stopping a job or downloading its results.

VMTKLab CFD on the Cloud: User Manual

HOW TO PREPARE A MODEL FOR DGETTO SOLVER The mesh has to be prepared in order to provide a good spatial representation of the surface domain but it should be more coarse as possible. The accuracy will be guaranteed by the polynomial degree of the solver and nor by the number of elements of the model. Having said that, meshes with 10k/500k of elements are acceptable. See the MultiDGetto solver paragraph and relative scientific publications for more information.

MULTIDGETTO SOLVER FEATURES

Robustness in convection dominated flow regimens is guaranteed by a fully implicit discontinuous Galerkin discretisation based on an artificial compressibility flux. We rely on a Godunov-like flux based on the solution of the Riemann problem associated with a local artificial compressibility perturbation of the equations. Artificial compressibility is introduced only for defining the inviscid numerical flux at interfaces, therefore resulting in a consistent discretization of the incompressible Navier-Stokes equations, irrespectively of the amount of artificial compressibility introduced (doi:10.1016/j.jcp.2006.03.006). BR2 dG discretisation of the viscous flux and physical frame orthonormal basis functions for increased robustness over distorted, curved and stretched mesh elements (doi:10.1016/j.jcp.2011.08.018).

Fully coupled monolithic solution strategy based on a FGMRES solver with h-multigrid preconditioning. The coarse grid levels are explicitly built (geometric multigrid) by recursive agglomeration of the fine grid, the generation of the coarse grids is robust and fully automated. Intergrid transfer operators (the restriction and prolongation operators) are explicitly computed based on L2 orthogonal projections. The resulting V-cicle is effective (good convergence factor are observed) with a single pre and post- smoothing iteration(doi:10.1142/S0218202514400028). Fast inexact-Newton continuation strategy for steady state initialisation of transient computations.

VMTKLab CFD on the Cloud: User Manual

VMTKLab CFD on the Cloud: User Manual

At least second order accurate in both space and time. It is possible to increase the spatial accuracy arbitrarily by increasing the polynomial degree of the dG discretisation. Adaptive BDF2 time marching strategy optimised for pulsatile computations with fast transients. Unstructured hybrid grids with tetrahedral, prismatic, hexahedral and prismatic elements with possibly curved edges (up to second order) are supported. Accuracy on curved elements meshes is maintained thanks to the use of polynomial spaces defined in the physical frame (doi:10.1007/s10915-011-9566-3). The solution strategy is scalable and the number of processes can be defined by the user based on a cost versus computation-time trade-off.