Concept lab · 01

Same traffic. Different queues.

Compare sticky sessions with stateless round-robin routing. Turn up one client’s traffic and watch where spare server capacity goes unused.

Clients12Servers4Requests per tick12Capacity per server / tick4Traffic from client 1 (%)60Workload seed

0.0 / 60.0 s · 0 requests sent

Replay position

Sticky sessions

Each client stays on its assigned server.

0 completed0 queued

Server 10 / 4 served · 0 queued

Server 20 / 4 served · 0 queued

Server 30 / 4 served · 0 queued

Server 40 / 4 served · 0 queued

Bars show capacity used in the latest tick.

Stateless round-robin

Each request goes to the next server in turn.

0 completed0 queued

Server 10 / 4 served · 0 queued

Server 20 / 4 served · 0 queued

Server 30 / 4 served · 0 queued

Server 40 / 4 served · 0 queued

Bars show capacity used in the latest tick.

Throughput requests / second · sampled every 100 ms

Sticky: 0 req/s · Round-robin: 0 req/s

What this model assumes

Both pools receive the same requests, with identical processing costs and server capacity. A tick is 100 ms. Client 1 generates the selected share of traffic; the rest is distributed among other clients by the seed. Sticky assignments use client number modulo server count. Arrivals enter queues before servers process up to their capacity each tick.

This illustrates load distribution, not measured MCP performance. Network latency, connection costs, shared storage, caching, failures, and session migration are omitted. Stateless routing assumes any server can handle any request. Queues are unbounded for this 60-second run; neither strategy drops requests.

Changing a parameter resets the run. The seed and parameters reproduce the same result at any replay position. Playback starts paused and pauses its clock in background tabs.