Ventana software support forum

The model that I've built is quite large and time consuming to run on my laptop. My university has High Performance Computing (HPC) facility centre that applies a compute cluster to computational problems. Interacting with an HPC cluster is typically done by typing commands into the Unix shell (usually BASH or the Bourne Again Shell). Typing-based interfaces are called Command-Line Interfaces, or CLI, to distinguish it from a Graphical User Interface, or GUI, which most people now use. More information about the facility can be found here: https://www.newcastle.edu.au/research-a ... -computing

I was wondering if such facility can be applied to my situation with Vensim software? Any different suggestions for accelerating my simulation runs will be much appreciated.

What do you want out of the HPC? Sensitivity analysis? Optimization? Faster single runs?

tomfid wrote: ↑Tue Feb 11, 2020 1:35 am What do you want out of the HPC? Sensitivity analysis? Optimization? Faster single runs?

I'm on optimisation (calibration) stage and it's taking long hrs.

We have a solution for this underway, but it's currently offline due to x64 upgrade turbulence. It's coming though.

Unfortunately calibration optimization is the worst case for performance - it doesn't parallelize easily. Sensitivity and MCMC will benefit much more.

tomfid wrote: ↑Tue Feb 11, 2020 1:53 am We have a solution for this underway, but it's currently offline due to x64 upgrade turbulence. It's coming though.

Unfortunately calibration optimization is the worst case for performance - it doesn't parallelize easily. Sensitivity and MCMC will benefit much more.

Thanks for informing me of the upgrade. The easiest way for now would be using a personal computer which is with more powerful specifications. However, since I purchased the license personally for myself and I've been using it on my personal laptop, I was wondering if I could use my license on another computer temporarily, and how?

Your license permits 1 user x 2 computers or multiple users on one computer, so that's OK.

A full HPC cluster may not be needed when the time comes - many-core workstations are now widely available and not particularly expensive.

One of the biggest speed improvements comes when you compile models, but this is only available to DSS users (your licence is Pro). Models can easily run 10 times faster when compiled.

There are other things you can do to speed things up. If it's a large model and you do not have a fast hard drive, you can use a savelist to only save those parameters of interest.

Also take a look at the following. It contains a few tips to optimize models.
https://metasd.com/2011/01/optimizing-vensim-models/

Ah, yes - that's the first thing to try. The speedup is structure-dependent, but I've never seen less than 2x and 5x is common, 10x possible if there's a lot of array structure.

Another thing to check is to be sure that any nondynamic computations are designated as INITIAL(). This is especially important for some of the GET DATA functions, like GET DATA LAST TIME, which can be slow.