1 of 12

Globus Compute Past and Future

Josh Bryan – josh@globus.org

August 28, 2025

ABSTRACT SUBMITTED:

The advanced computing playing field is fundamentally non-level: a small fraction of researchers have access to substantial resources—with respect to both capacity and capability—that allow them to easily scale their work from proof-of concept to large-scale experiments; but the majority are constrained by a lack of access and the advanced skills required to move from small- to ultra-scale.

A number of efforts, such as the recently announced ACCESS project, are making it easier for disadvantaged groups to compute at scale, but the technical hurdles remain. We will describe and demonstrate how a novel implementation of functions-as-a-service enables even those with only rudimentary programming knowledge to easily compute at any scale, from a laptop to a supercomputer.

2 of 12

Globus Compute

What have we been up to?

3 of 12

High Assurance Support

Stronger authentication requirements

Recent In-Session Authentication
Specific Identity Providers
MFA requirements

Support usage with CUI / PHI
Audit Logging
Cloud services operational controls

4 of 12

Administrative Console

5 of 12

Flows Action Provider v3

Better support for multi-user endpoints
Improved serialization between AP and endpoint
Support for resource specification

6 of 12

Globus SDK / CLI integration

Support for web API calls directly in Globus SDK
Endpoint management through SDK / CLI
Better authentication flow handling through Globus App
…. Lots of bug and papercut fixes

7 of 12

Globus Compute

What do we have planned?

8 of 12

Templates for All

By default, single user endpoints will support templating
Unified execution model between single and multi-user endpoints
Support for more complex template structures

DRY: Importing templates
More context info for template administrators

9 of 12

Better documentation

Better examples and tutorials
Better organization
Simplification of concepts

10 of 12

And more

Pluggable serialization support for better integration with Parsl and less mismatched version issues
Some task lifecycle features

Clearing task queues
Better recovery from restarts

Console views for single user endpoints