Gladier builds on a science services framework, in the form of Globus Auth, Transfer, Search, Groups, and Flows, plus Comptue function-as-a-service. These services provide a reliable, secure, and high-performance substrate to access and manage data and computing resources. Here we highlight several of these services and describe how they are used to create Gladier deployments.
Globus provides a collection of data services built for science including: Globus Auth, Transfer, Search, Groups, and Flows, and Compute to enable distributed function-as-a-service execution. Globus Services are highly reliable, professionally operated cloud-hosted services that support the work of over 150,000 researchers worldwide as foundational capabilities for scientific applications and workflows; using them greatly reduces the burden on local systems, administrators, and programmers.
Flows addresses the problem of securely and reliably automating sequences of data management tasks that may span locations, storage systems, administrative domains, and timescales, and integrate both mechanical and human inputs. Client libraries deployed on Globus endpoints and other sources enable the detection of events and invocation of a flow. The Flows service manages execution of user-supplied automation flows either manually or as a result of data events, and the invocation of actions from those automation flows, including actions provided by Globus endpoints and services. The service is extensible via the definition of new events and actions to meet the needs of specific communities.
Globus Compute is a function-as-a-service platform that implements a federated compute substrate, enabling computation to be registered as Python functions and invocations to be dispatched to remote computers for execution. The service provides a single point-of-contact, supporting function registration, sharing, and discovery as well as reliable and secure execution on connected endpoints. The compute endpoint software, built on Parsl, allows functions to be executed in containers and for resources to be dynamically provisioned on cloud and cluster systems. These compute endpoints provide serverless capabilities whereby researchers fire-and-forget tasks that are dynamically allocated across the supercomputer using an opportunistic backfill queue to utilize spare capacity.
Globus Queues and Triggers
Queues provides a reliable, cloud-based mechanism to manage and store events. The Queues service allows users to provision a dedicated queue for their instrument. Clients can then raise events to the queue using HTTP POST requests where they will be maintained until a subscriber consumes them. This enables experimental facilities and instruments to raise events as data are created without requiring heavy-weight installations on the edge device.
The Triggers service provides a cloud-based consumer of Queues events. Users can configure a Trigger to monitor a queue and initiate Flows as events are received. To create a trigger one defines:
An event queue for the trigger to monitor
A condition for when the trigger will fire
An action to perform when the condition is met (e.g., a flow uuid)
A template to create an input JSON document for the action. This often includes default values.
The combination of the Queues and Triggers services simplifies creating new Gladier deployments.
Transfer implements a location-agnostic data substrate that enables data to be accessed, shared, and moved among disparate storage systems, including at instruments, supercomputers, and on data services. Globus Transfer allows users and applications to modify data access permissions on remote storage systems and to move data reliably and securely between systems via a single API.
Auth Auth allows for users to delegate permissions for clients to access services in the Gladier architecture, and for services to access other services on their behalf as well. For example, it allows Globus Flows to manage Globus Transfers and to execute analyses via Compute on systems accessible only to the user.