Managing distributed scientific workflows with Globus

Scientific workflows increasingly span remote computing resources, from local desktops and scientific instruments to supercomputers, clouds, and AI accelerators. This distribution is driven by the nature of modern data-driven research and the availability of specialized computing hardware. Distribution creates new opportunities to improve performance and efficiency by exploiting resource heterogeneity and locality; however, it also creates new challenges related to portability and security. In this chapter, we describe Globus, a platform designed to tackle these challenges via a hybrid model in which cloud services securely manage the remote execution of arbitrary research activities. We describe how Globus Flows, a cloud-hosted workflow platform, combined with Globus Compute and Globus Transfer, enables researchers to define and execute workflows across diverse distributed computing resources. We present several example applications in real-time instrument analysis, simulation campaigns, and distributed model training that demonstrate how Globus addresses challenges in real-world scenarios.