The Production Services team, as part of the Fabric of Capacity Services (FoCaS), maintains services essential to delivering science on the Open Science Grid. The Compute Management services allow for sharing of computing resources, provisioning pools of computing resources, job execution management. Data Management services provide for serving and pre-placing data in support of workflows. Additionally, services such as identity management, monitoring, and accounting enhance and secure the fabric.
The compute management services allow the science and engineering community to turn the raw computing capacity found in clusters across the nation’s universities and labs into effective capacity for science.
The computing services provided by PATh include:
- Submission endpoints for providing researchers access to the shared compute power enabled by the OSG. This includes the centrally-managed OSG Connect, targeting PI-driven groups.
- Overlay resource pools for aggregating resources from the distributed infrastructure on behalf of an organization.
- Resource Provisioning for acquiring resources – from hosted CEs, XD machines, or cloud capacity – into the resource pools.
- Compute Entry points for contributing resources via a scheduler (such as SLURM, HTCondor, or PBS) on a campus cluster into a resource pool, using the HTCondor-CE software (a special configuration of HTCSS).
PATh delivers data management services in order to support the workflows running on the national infrastructure. Services include:
Distributed data caches provide on-demand data movement to jobs. A data cache is located near (in network terms) compute resources and manages the usage of WAN bandwidth to mitigate latency impact. For workloads with high locality of reference, caches allow for data reuse and reduction of total WAN bandwidth and load on data origins.
Data placement services allowing an organization to maintain a catalog of its files, organize files into datasets, establish placement policies, and orchestrate transfers with an external service.
Research software distribution allowing organizations to publish software or containers and, within minutes, have them propagated across the compute federation.