Inductiva API v0.6 is out!

We are happy to announce that Inductiva API v 0.6 is now available.

This version brings a number of improvements in performance and usability based on feedback received from our users.

Ability to submit and download very large files

It is now possible to call simulators and pass very large input files (2Gb+), such as, for example, when you need to pass a very dense mesh to OpenFOAM. We also improved the mechanism for downloading the results of simulation, which is now more efficient and capable of dealing with larger files. This addresses one of the top requests from our users: thank you for your feedback!

Optimized Logging Management

We changed the way we are collecting real-time logging from our simulation execution environments, to reduce latency and improve overall performance of running simulations that send very large amounts of information to stdout. The new default behavior of the API is to not collect remote stdout logs in real-time, making sure the performance of simulators is preserved. When needed, such as in development time, users can always optionally turn on streaming of remote logging. In any case, nothing is lost: all logging information is still available for downloading, but just after the execution of the simulation.

Smarter Scaling Up/Down Policy for Elastic Machine Groups

We improved the resource management policy of our Elastic Machine Groups to make use of information about the number of tasks waiting on queues as triggers for scaling up and down operations. This is an improvement with respect to the default scaling policy provided by GCP, which uses internal computational load of the VM to trigger changes in the elastic pool of resources. Our new scaling policy is able to process typical loads coming from simulation workflows faster than the default one.

CaNS: one additional CFD simulator is now available

We added support for CaNS (Canonical Navier-Stokes), a massively-parallel numerical simulator of fluid flows. This is an important addition to the CFD simulator options we have available, and we are working directly with the CaNS developers to ensure it is operating at the best possible level of performance.

Know your limits

We made it easy for users to know how much of their quota they are using and how much is left. When using Inductiva’s Command Line Interface, you can simply issue inductiva quotas list to get a detailed list of all the quotas together with their current usage. You can also programmatically obtain quota information from your Python script using command inductiva.users.get_quotas(). This is very helpful when attempting to run large jobs that may request launching more machines than the limit you have in your quota. These mechanisms will allow you to request more resources only when you have enough quota for that, and thus avoid errors arising from exceeding your resource allocation limits.

Additional Improvements

We made important improvements on other aspects of the broader Inductiva ecosystem.

A big welcome to Inductiva Benchmarks for Open-Source Simulators

As part of our effort to continuously improve the supercomputing experience of all our users, we are proud to be officially releasing Inductiva Benchmarks for Open-Source Simulators. These benchmarks compare the time and cost of running different use cases for several simulation packages over all VM types that are available via Inductiva API. This data helps understand how well certain simulation packages scale their performance with the number of vCPU / threads it is run on, or if there are significant performance variations depending on whether they are running on different VM families (ex Intel-based vs AMD-based). We will be continuously adding more use cases for more simulation packages, so you should expect more news on this soon.

Kutu keeps growing!

Besides CaNS, which is already available via API, we added to Kutu, our publicly-available repository of Docker containers, AMR-Wind, a massively parallel, block-structured adaptive-mesh, incompressible flow solver for wind turbine and wind farm simulations, jointly developed by Lawrence Berkeley National Laboratory, National Renewable Energy Laboratory, and Sandia National Laboratories.

Kutu now contains Docker images for 12 different simulators.

Inductiva API Tutorial web site

We now have a dedicated site for Inductiva API Tutorials, where we will be posting information on how to use the API for a number of different practical use cases. The first tutorial is already available and it guides the user through the process of generating data using a Smoothed Particle Hydrodynamics simulator for training a GNN model to predict complex fluid dynamics.

What’s next on the pipeline?

We keep listening to our users, so in the next release you should expect a few more additions based on valuable feedback we have been receiving. So:

We want to make it easier for you to start using the API. Very soon you will be able to login on the API using your existing credentials from Google, GitHub and others.
We don’t want you to have to run all simulations remotely. Sometimes, it just does not pay off, especially when you have a pretty fast local CPU. We are working on letting you run the API locally if you wish, only submitting your jobs to run on remote machines when you really need to scale.
We are making a big effort to make more and more simulators available, by adding more images to Kutu, but we know that is not enough. So, soon, we will provide you with a mechanism for running arbitrary containers, your own containers.

Stay tuned… and simulate a lot!

Our Insights