Ctrl+K

Computational Infrastructure

Computational Infrastructure#

The Inductiva API serves as a direct intermediary, bridging the gap between users and the complex landscape of computational resources. It streamlines the process of managing and allocating computational workloads and simulation tasks, by facilitating access to an expansive selection of computing options.

This guide will detail how the API simplifies the orchestration of your simulations and introduce you to the various computational options currently available to you, as well as preview exciting additions planned for future releases!

Computational Resource Management#

The Inductiva API acts as an abstraction layer that enables you to access a wide array of computational resources provided by a number of different players through a unified Python code. These resources could be from cloud providers, bare-metal hardware rentals, standard high-performance computing (HPC) solutions commonly used in academia, or even on-premise hardware for those with their own computing infrastructure. The API serves as a unifying interface atop all these varied resources, facilitating access to computational solutions of varying scales, prices, and performance levels and helping you select the optimal resource for your needs, all through straightforward Python scripting from your laptop.

From the server side, Inductiva manages your computational workload — be it one or several simulations. It allocates this workload to the appropriate computational resource dedicated to you, handles the orchestration of the simulation, and then returns the results back to you.

Available Computational Resources#

In the current release, version 0.5, Inductiva supports dispatching computational workloads to the Google Cloud Platform (GCP). This means that the simulations initiated through our API are executed on one or more virtual machines (VMs) hosted on GCP.

There are two families of Virtual Machines (VMs) made available by Inductiva on Google Cloud Platform (GCP):

The VMs within these families are categorized into three types, based on the RAM-to-vCPU ratio:

  • highcpu - Offers 2 GB of RAM per vCPU, suited for CPU-intensive tasks.

  • standard - Provides a balanced 4 GB of RAM per vCPU, ideal for general-purpose use.

  • highmem - Equipped with 8 GB of RAM per vCPU, designed for memory-intensive applications.

These configurations allow for the customization of MachineGroups, ElasticMachineGroups, and MPIClusters to match your computational needs.

Here’s an example of how you can start a MachineGroup with robust “c3d-standard-60” general machines:

import inductiva

# Initialize a MachineGroup with two "c3d-standard-60" machines
machine = inductiva.resources.MachineGroup(
    machine_type="c3d-standard-60",
    num_machines=2,
)

# Start the MachineGroup
machine.start()

Naturally, the cost associated with each machine type varies, and it’s possible to accrue significant expenses if a large number of VMs are initiated! To protect you from inadvertently spinning up too many resources, the API imposes certain limitations on the quantity and types of machines that you can launch. For details on these limitations, please consult the User Quotas we put in place through the current version of the API.

See also

Learn how to manage your computational resources through Inductiva’s Command Line Interface

Upcoming Computational Resources#

In future versions of the API, we plan to expand the range of computational resources you can access through the API. Here’s a sneak peek at what is on our development roadmap:

  • Inductiva Compute Engine (ICE): We are developing our own infrastructure to offer both CPU-based and high-performance GPU-based VMs. These will be particularly beneficial for simulator packages requiring extensive computational power.

  • Bring your own Cloud (BYOC): This feature will enable you to use your own cloud accounts, starting with GCP, to run simulation jobs.

  • Bring your own Hardware (BYOH): With BYOH, you’ll be able to integrate the API directly with your existing hardware, whether it’s on-premise or leased, maximizing the resources you already invested in.

  • High-performance Computing (HPC): We are collaborating with HPC providers to allow redirection of certain simulation loads to large HPC clusters you can access.

Stay tuned for these updates, which will be rolled out in the upcoming releases!

Contents