Abhijit Mazumder

LAB 02 / ARRIVAL-MODEL SIMULATOR

Same server.
Different arrivals.

Watch a closed-loop client wait while an open-loop schedule continues creating demand. The difference becomes visible the moment the server stalls.

ILLUSTRATIVE SIMULATION / NOT MEASURED

What arrives while the server is stalled?

The server is identical in both lanes. Only the rule that creates the next request changes.

Ready

Playback expands the selected stall to 4.5 seconds on screen. Simulation time remains actual.

In the closed loop, clients wait and stop creating additional requests during the stall. In the open loop, scheduled arrivals continue and form a queue.

At 50 requests per second, a 1,000 millisecond stall schedules 50 additional open-loop requests and zero additional closed-loop requests.

CLOSED LOOP0new requests during stall
OPEN LOOP50scheduled during stall
OPEN-LOOP RECOVERY1.0 sto drain the illustrative queue

Claim boundary. This deterministic fluid simulation explains arrival semantics. It is not a vLLM run, production trace, or latency prediction.

Read the method note →

WHAT IT SHOWS

Arrival semantics change the queue.

A closed-loop client creates its next request only after a response. An open-loop schedule keeps creating requests against wall-clock time. During a shared server stall, the second model preserves demand that the first never generates.

WHAT IT DOES NOT SHOW

This is not a latency prediction.

The animation uses a deterministic fluid queue and a simplified global stall. It does not model vLLM scheduling, batching, GPUs, network delay, or a production arrival trace. Those belong in the measured experiment.