Data infrastructure
for Physical AI.

Built into BotRanch.

Physical AI workflows depend on connected scene assets, simulation data and field records. BotRanch includes an internal shared data layer for asset caching, simulation shards and replay access. An RDMA memory pool and POSIX and key-value interfaces support data access, with persistence and availability matched to the task.

AI TRAININGINFERENCESIMULATION Shared data layerSHARED MEMORY · RDMA NVMe / NODE 01NVMe / NODE 02NVMe / NODE 03
Data architecture schematic

Share the data
each workflow needs.

High-performance data infrastructure is an internal BotRanch capability. Evaluate access paths for simulation assets, training data, model weights, and checkpoints, with interfaces and deployment scope agreed for each scenario.

RDMA data path

Connect nodes through remote direct memory access for hot data and metadata.

NVMe capacity tier

Serve hot data from memory and extend capacity with local NVMe.

POSIX / KV interfaces

File and key-value interfaces support different data models.

Data protection options

Choose cache acceleration, persistent storage, or high-availability deployment according to the task.

Choose the protection
your workload needs.

Deployment of BotRanch’s internal data layer depends on whether data can be rebuilt, must persist, or needs automatic failover. Confirm the configuration during technical evaluation.

A shared acceleration layer for rebuildable data.

For scene assets, model weights, training caches, and temporary compute data. Keep source data in object storage, NAS, or another system so it can be reloaded after restart.

NNode count
POSIX / KVAccess interfaces
RebuildableData role
  • RDMA memory pool and one-sided reads
  • NVMe capacity tier
  • No write-ahead log
  • Reload source data after restart
View evaluation approach

Cache acceleration does not guarantee persistence. Retain recoverable source data; a cache should not be its only copy.

CapabilityCache accelerationPersistent storageHigh-availability deployment
RDMA memory pool / NVMe tierIncludedIncludedIncluded
POSIX / KV interfacesIncludedIncludedIncluded
Write-ahead log and checkpoint—IncludedIncluded
Asynchronous standby & failover——Included
Node countNN2N

These are architectural options for the data layer. Interface compatibility, persistence semantics, and recovery objectives are determined by the deployment environment and technical evaluation.

Evaluate the architecture
with real workloads.

Evaluate actual data, access patterns, and deployment environments. Agree on configurations and acceptance criteria before testing, and record comparable results to inform deployment decisions.

Data access

Use simulation assets, models, or checkpoints to assess throughput, latency, and metadata handling.

Parallel scaling

Compare node configurations and shared workloads, recording resource use, scaling efficiency, and operating costs.

Recovery and availability

Evaluate persistence semantics, failover, recovery time (RTO), and the data-loss window (RPO) for the chosen configuration.

Define the technical scope
from the real task.

How do I choose a data deployment option?

Evaluate cache acceleration for reloadable data, persistent storage for data that must survive restarts, and high-availability deployment for services needing automatic failover. Define recovery time and acceptable data-loss bounds.

How should we evaluate the data layer?

Choose a real simulation-asset, checkpoint, model-loading, or key-value workload. Compare task throughput, waiting time, resource use, and cost while recording test conditions.

What role does the data layer play in BotRanch?

High-performance data infrastructure is an internal BotRanch capability for asset caching, simulation shard sharing, and replay access. Workflows connect scenes, tasks, policies, and evidence. Data configurations follow the pilot scope, while safety control remains with field controllers.

Build your next
Physical AI workflow.

Discuss BotRanch demonstrations, deployment requirements and partnership opportunities.

Get in touch