DatacenterExplorer

What it is. An Unreal Engine plugin that ingests a hierarchical JSON description of a data-center (cluster → rack → server → component) and brings the hierarchy to life as Unreal actors you can fly through, filter, recolor, and dynamically update. Part of the ExaDigiT framework — ORNL’s project for building digital twins of exascale supercomputers.

Why. A digital twin only earns its name if you can actually see the system, query it interactively, and watch live telemetry move across it. DatacenterExplorer is the visualization layer of that pipeline: hierarchy in, navigable scene out.

What it does

• Blueprint-first. Every primary capability is reachable from Blueprints, not just C++.
• JSON-driven hierarchies. Nested JSON files describe nodes, positions, references to sub-hierarchies, and visibility — parsed with template caching so repeated structures (a rack with N identical servers) don’t re-parse.
• Strategy-based spawning. A HierarchyStrategist picks how the hierarchy is walked (spawn / visual / destroy strategies), generates commands for the HierarchyManager, and the manager processes the queue with adaptive frame-rate management — so a 100k-component cluster doesn’t tank the editor.
• Flexible visualization. Material strategies recolor nodes based on property filters (name == *gpu* → red; temperature > 80 → orange). Rules are regex-capable and combine cleanly.

How the pipeline works

Parse → HierarchyReader loads JSON files into a HierarchyNode tree
Strategy → HierarchyStrategist picks spawn / visual / destroy
Commands → Strategy generates commands for HierarchyManager
Execute → Manager processes the queue with adaptive frame management
Visualize → Actors spawn with materials / properties from node data

The split between parsing (HierarchyReader), strategy (HierarchyStrategist), execution (HierarchyManager), and filtering (HierarchyNodeFilters) is the load-bearing abstraction — a new visualization (heatmap by temperature; mesh-swap based on failure state; show-only-this-rack) is a new strategy, not a plugin rewrite.

Example configs in the box

The repo ships five complete example hierarchies, covering more than three orders of magnitude in scale:

Example	Scale	Approx. actors
L01_tiny_micro-ex	demonstration	~50
L02_small_ex-unit	single unit	~1,000
L03_medium_marconi	mid-size cluster	~20,000
L04_large_summit	ORNL’s Summit	~100,000
L05_huge_frontier	ORNL’s Frontier (exascale flagship)	~150,000+

Each example is a top-level cluster.json referencing component JSONs (cabinets, blades, CPUs, GPUs, memory, PDUs, rectifiers, network groups, environmental controllers). At Frontier scale custom strategies become necessary — the plugin is fast, but spawning 150,000 actors honestly is still 150,000 actors.

Live telemetry via DataPipeline

DatacenterExplorer pairs with the sister plugin DataPipeline to wire live telemetry (temperatures, power, utilization) into the scene at runtime. The hierarchy is the what’s where; DataPipeline is the what’s happening right now.

Use cases

• Data-center infrastructure visualization
• HPC system monitoring & digital twins
• Interactive architecture exploration
• Real-time component-status visualization