GlideinWMS

GlideinWMS, short for Glide-in Workload Management System, is a grid computing middleware that provides a dynamic, pilot-based approach to running work across distributed resources. It creates virtual pools of compute capacity by submitting pilot jobs, called glideins, to remote clusters managed by various local resource managers. The pilots then pull user work from a central queue and execute it within the glidein environment, enabling a single virtual workload space that spans multiple sites.

The architecture centers on three elements: the GlideinWMS Frontend, the GlideinWMS Factory, and the glidein pilots. The Frontend receives workload from a Virtual Organization (VO) and determines how many glideins are needed to satisfy the queue. The Factory is responsible for provisioning glideins on candidate sites, interfacing with the local resource managers (such as HTCondor, Slurm, PBS, or LSF). The running glideins register with the central components and execute user jobs on the remote resources. In typical deployments, the glideins are based on the HTCondor glidein model, though GlideinWMS can integrate with multiple LRMS backends.

Workflow-wise, a VO submits jobs to the GlideinWMS frontend, which requests glideins from the factory. Once glideins are active, they connect to the VO’s job pool, fetch suitable tasks, start execution on the remote resource, and report results back to the VO. This setup enables dynamic resource provisioning, improved utilization, and workload sharing across sites without requiring permanent allocation of dedicated clusters.

GlideinWMS is widely used in large-scale grid environments, notably within the Open Science Grid and European grid infrastructures, to support diverse scientific workloads while maintaining a centralized, VO-focused control plane.