メインコンテンツまでスキップ
This page is not translated yet. 日本語 is switched on so the navigation and URLs can be tested; the text below is the English original. Use the language selector in the header to switch back, or see the documentation home.

Intelligent Resource Automation

Intelligent Resource Automation (IRA) is a group of platform features that let Rescale pick the right hardware, region, and license window for a job instead of pinning every submission to a fixed configuration. IRA addresses three recurring constraints in cloud HPC: software licenses are finite, cloud capacity varies by region and hour, and specific geographies carry compliance requirements. This page covers the three IRA features available today: License Queuing, Region Load Balancing, and Coretype Sets.

When to use Intelligent Resource Automation​

Reach for IRA features when any of the following is true.

  • Your jobs fail because a license feature is temporarily unavailable. Use License Queuing to hold the job until tokens free up.
  • Your preferred Coretype is capacity-constrained in a region you use. Use Coretype Sets to define an ordered fallback list.
  • You want jobs distributed across cloud provider regions based on availability and region health, rather than pinned to one region by default. Use Region Load Balancing.
  • You have compliance requirements (for example, ITAR) that restrict which regions a job can run in. Combine Region Load Balancing with your compliance level so non-compliant regions are excluded automatically.

IRA features are configured at the user, workspace, or partner level by administrators. Many of them require working with Rescale Support for initial setup.

License Queuing​

License Queuing holds a job until the license features and token or seat count it requires are available from the requested license server, then launches it. It is available to all Rescale users and works with Rescale-hosted licenses as well as a company license proxy, with or without a VPN.

Benefits:

  • Fewer job failures caused by licenses being temporarily unavailable.
  • Simpler job configuration via configurable license queuing controls.
  • A predefined token-usage formula for supported software, so you do not have to calculate feature counts by hand.

To use it, configure the license feature tokens your job requires at submission time. Rescale tries to obtain the tokens before launching the job and retries automatically if they are not yet available.

Usage guidelines
  • License Queuing does not work for customers who use a VPN but no company license proxy.
  • License Queuing is not recommended for Design of Experiments jobs, because DOE jobs fan out into many parallel children with simultaneous token demand. See DOE and parameter sweeps.
  • If a license feature is not found on the license server, the job will continue to run.

For broader context, see the Running a Basic Job topic for job-level license configuration, the License Hosting topic for Rescale-hosted licenses, and the License Proxy topic for connecting an on-premises license server to Rescale.

Region Load Balancing​

Region Load Balancing distributes compute and storage across cloud service provider (CSP) regions based on real-time capacity and region health signals. It is on for all users by default and helps avoid two concrete failure modes.

  • Capacity exhaustion. If a region is low on capacity, jobs wait longer or lose hardware mid-execution. Load balancing sends jobs to regions with headroom.
  • Region degradation. If a region is suffering operational issues, jobs submitted there run slower or fail. Load balancing avoids regions with poor health scores.

By default, Rescale picks from a set of regions using dynamically calculated Capacity Health Scores. You can override the default by setting region preferences at the user, workspace, or partner level; Rescale honors those preferences above its own ranking.

Two constraints worth calling out:

  • Compliance. If you have compliance requirements such as ITAR, set the compliance level so Region Load Balancing never routes compute to non-compliant regions.
  • High-Performance Storage. If a job or desktop launches with HPS attached, the job is pinned to the region where the HPS cluster already runs. Region Load Balancing will not move it elsewhere.

To configure region load balancing for your organization, contact Rescale Support.

Coretype Sets​

A Coretype Set is an ordered list of coretype and priority pairs Rescale tries in turn, so a job starts even when your first choice has no capacity. IRA is what acts on the Set at launch, but the Set itself is a hardware concept, so it is documented with the hardware: Coretype Sets.

  • Hardware for Coretype Collections and Compute Optimizer, which also shape hardware selection.
  • DOE and parameter sweeps for why License Queuing is not recommended with DOE.
  • Jobs for the underlying hardware and license configuration.