Problems / LLM serving, hard
A KV-cache tier sits in front of prefill, but only 10% of requests reuse a cached prefix: the system prompt changes per user, so prefixes never match.
Free to open. Runs in your browser, no sign-up.
Success ≥ 99%, p99 ≤ 5 s.
260 requests a second for 30 seconds: 100% request (reads).
What each component does in this design, and what calls it. The simulator mimics each one from these settings.
| Component | Its job here | Settings |
|---|---|---|
| Load balancerLoad balancer | Traffic enters here.Calls AI app.Layer 4 balancers route by IP and port; layer 7 balancers read HTTP and can route by path or header. | – |
| AI appApp service | Load balancer forwards to it.Calls KV cache tier.Little's law: throughput ≈ concurrency ÷ latency. 200 slots at 50 ms serve about 4,000 req/s. | 3 replicas × 500 slots · timeout 30 s |
| KV cache tierKV cache tier | AI app calls it.Calls Prefill GPU pool.Reusing the KV cache of a shared prefix (system prompt, chat history) skips most of the prefill work. | 2000 slots · 10% hit rate |
| Prefill GPU poolPrefill GPU pool | KV cache tier calls it.Splitting prefill (reading the prompt) from decode (writing the answer) onto separate GPUs stops long prompts from stalling everyone else’s output. | 40 slots |
Try it first. Open one hint at a time if you get stuck.
Each miss is a full prefill.
Shared prefixes must come first and be byte-identical.
Raise the KV cache hit rate by putting the shared prompt first.