System Design Simulator

{{ n.type==='source' ? 'source' : n.category }} {{ n.name }} ×{{ n.replicas }}

Total{{ fmt(n.readRps + n.writeRps) }} rps

Reads{{ fmt(n.readRps) }} rps

Writes{{ fmt(n.writeRps) }} rps

Load{{ fmt(n.sim.load) }} rps

Latency p99{{ n.sim.p99.toFixed(0) }}ms

Utilization {{ (n.sim.util*100).toFixed(0) }}%

Drag components from the left to design your system,
then connect them by dragging between the ports.

{{ hint }}

System Metrics

Latency p50

{{ global.p50.toFixed(0) }}ms

Latency p99

{{ global.p99.toFixed(0) }}ms

Served

{{ fmt(global.served) }}rps

Error rate

{{ (global.errorRate*100).toFixed(2) }}%

Availability

{{ (global.availability*100).toFixed(3) }}%

Est. cost

${{ global.cost }}/hr

Max sustainable load (before saturation)

{{ fmt(global.maxLoad) }}rps ({{ global.headroom.toFixed(2) }}× headroom)

Bottlenecks & Alerts {{ alerts.length }}

{{ a.title }} — {{ a.msg }}

{{ testResult.title }}

{{ l.name }}

Metric Stream

{{ ev.t }} {{ ev.m }}

Edit: {{ selected.name }}

Name

Select a component on the canvas to tune its parameters, or drag a "kill" toggle to inject a failure.

Workload

Incoming (from source)

Request load (rps)

Latency SLO (ms)

Model: M/M/1 queueing per node (utilization ρ = load ÷ capacity·replicas). Caches are read-through — they serve hit-rate% locally and forward only misses to the origin wired behind them. Routing & queue nodes split load across children; other nodes call each dependency at full rate. Replicas add redundant availability 1−(1−a)ⁿ. Percentiles are analytic estimates, not a packet-level trace.