Locales: representing architectural localityΒΆ

In Chapel, a locale is a subset of the target architecture that can be used to control and reason about affinity for the sake of performance and scalability. Things that are co-located within a single locale are close to each other in the system and can therefore interact with one another relatively cheaply. Things that are in distinct locales can still interact with each other in the same ways, simply at greater expense.

As an example, a task running on a given locale can refer to variables via standard lexical scoping rules regardless of which locale the variables are stored on. However, a variable stored on the same locale as the task can be accessed more cheaply than if it were stored on some other (remote) locale.

For most target architectures, a locale is equivalent to a compute node. For example, a program that is specified to execute on 16 locales will typically be executed using 16 compute nodes.

All locales have the ability to run tasks and to store variables. This means that a locale will contain processors and memory (or have access to processors and/or memory that it can use by proxy).

Locales may themselves contain locales (sub-locales), but this capability is not yet widely used, so it is not yet covered in this guide.