AWS Launches Managed Slurm Service PCS in Seoul Region to Minimize Operational Burden

Managed Slurm Cluster, AWS PCS

AWS Parallel Computing Service (PCS) is a managed service where AWS handles the operation of the Slurm scheduler for HPC clusters. Launched in 2024, it was introduced to the Seoul region in July 2026. Unlike the existing ParallelCluster, AWS performs patching, upgrades, and failure recovery for the head nodes (controllers) that customers previously managed directly, while compute nodes and data remain within the customer's account on EC2.

Zero-Downtime Operations and Reduced Operational Burden

The core value of PCS is zero-downtime upgrades that do not stop running jobs. Scheduler security patches and minor updates are applied automatically, and major upgrades proceed in-place without cluster regeneration. Compute nodes can update OS and drivers via rolling replacement. Through this, Shell stated that it reduced cluster management efforts by 75% when migrating its internal HPC environment. Controller availability is guaranteed by a monthly 99.9% SLA.

Cost Structure and TCO Perspective

PCS is a paid service, with controller fees and node management fees added on top of EC2 compute costs. Controller fees are charged fixedly for the time the cluster is in the ACTIVE state. For 20 nodes running continuously, management costs amount to approximately $2,157 per month, and the proportion of fixed controller costs increases even with intermittent usage. However, judgment should be made from a Total Cost of Ownership (TCO) perspective considering labor costs and failure losses incurred during direct operation, and prioritizing PCS review is recommended for new workloads.