RDMA data path
Connect nodes through remote direct memory access for hot data and metadata.
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.
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.
Connect nodes through remote direct memory access for hot data and metadata.
Serve hot data from memory and extend capacity with local NVMe.
File and key-value interfaces support different data models.
Choose cache acceleration, persistent storage, or high-availability deployment according to the task.
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.
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.
Cache acceleration does not guarantee persistence. Retain recoverable source data; a cache should not be its only copy.
| Capability | Cache acceleration | Persistent storage | High-availability deployment |
|---|---|---|---|
| RDMA memory pool / NVMe tier | Included | Included | Included |
| POSIX / KV interfaces | Included | Included | Included |
| Write-ahead log and checkpoint | — | Included | Included |
| Asynchronous standby & failover | — | — | Included |
| Node count | N | N | 2N |
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 actual data, access patterns, and deployment environments. Agree on configurations and acceptance criteria before testing, and record comparable results to inform deployment decisions.
Use simulation assets, models, or checkpoints to assess throughput, latency, and metadata handling.
Compare node configurations and shared workloads, recording resource use, scaling efficiency, and operating costs.
Evaluate persistence semantics, failover, recovery time (RTO), and the data-loss window (RPO) for the chosen configuration.
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.
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.
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.
Discuss BotRanch demonstrations, deployment requirements and partnership opportunities.