Capacity explorer
Build a workload scenario and compare it with measured results.
Loading benchmark measurements…
Expected traffic
Measured reference
Compare node sizes
Resources are per node. Every node stores a complete copy of the database. Leader RSS is sampled process memory during scans, including mapped database pages; it is not a heap-size requirement.
| Per-node resources | Point read p95 | Write p95 | Repeated search median | First full scan median | Evidence |
|---|
Combined workload checks
Longer rate checks
| Per-node resources | Operation | Offered / completed per second | p95 | Failures / drops |
|---|
How to use this comparison
People and agents become an estimated request rate using the activity values you enter. The table uses the next larger measured dataset; it does not interpolate a CPU-per-record formula.
Latency measurements include the test client’s network path. Point reads, writes and full scans were measured separately. Passing those checks does not prove that an arbitrary combined workload or burst will meet the same times.
The current MCP exposes contact search but no contact-by-ID tool. Agent-heavy workflows may need a much higher full-search share than this example.