HPC Migration — Control Tower

HPC Cluster Discovery

Source Cluster

☁

AWS Cloud

ParallelCluster on EC2

🏢

On-Premises

Hybrid / On-Prem HPC

Discover all clusters in regions (comma-separated, blank = all)

Cluster name

AWS region

Head node instance ID (recommended)

When the head node is running & SSM-managed, discovery reads the cluster config, software stack and identity directly from it.

📡

Scan a cluster to begin

Enter the cluster name, region and (ideally) the running head node ID, then Scan cluster. Discovery builds the PCS target blueprint, software/identity inventory, and storage map.

Migration pipeline — ParallelCluster → AWS PCS

Start migration

Source cluster

Target PCS cluster

Approver email

Optimise for

Pipeline stages

Live logsrefresh

no activity yet

Migration run history

Runs started from this console clear

Started	Source cluster	Target	Region	Optimise	Execution	Status

Migration documentation

Technical migration document

Deployment guide — ParallelCluster → AWS PCS

Connection

Discovery backend

If the account blocks public Function URLs, run discovery_proxy.py and set this to /api/discovery. Otherwise paste DiscoveryUrl from the stack outputs.

Discovery Lambda URL

Scanner Lambda URL (lists clusters across regions)

Migration Lambda URL

Logs Lambda URL

Approval API URL

Console token (ConsoleTokenSecretArn — aws secretsmanager get-secret-value)

Default region

Bedrock model ID (advisory)

What discovery captures

Jobs are not migrated — the cluster definition is. Discovery captures, in priority:

1. Config → PCS — Slurm partitions become compute node groups & queues (instance type, min/max, EFA, GPU).
2. Software & identity — compilers, MPI, CUDA, modules, users/LDAP — so jobs run identically.
3. Storage & networking — FSx/EFS mounts, VPC, EFA.
Workload (optional) — Slurm accounting for right-sizing & cost only.

Full fidelity (1 & 2) needs the head node running & SSM-managed; otherwise discovery returns the cloud-API view (api-only).