Problems / LLM serving, medium
One GPU pool with 64 slots serves interactive chat (30 requests a second) and a background agent that triages every ticket: 10 tasks a second, 8 model calls each. Nothing limits how many background tasks run at once.
Unlock it once for 15 credits; everything inside it is free (runs, changes, Submit, results, hints). Sign in with Google to start; new accounts get free credits.
Chat success ≥ 99% and p99 ≤ 2 s; background tasks still finish.
Like a coding problem, a solution has to pass every case, not just the one it was built for. Submit in the lab runs all 5; 2 of them are hidden until then.
40 requests a second for 45 seconds: 75% chat (reads), 25% background (writes).
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 Chat API, Background job API.Layer 4 balancers route by IP and port; layer 7 balancers read HTTP and can route by path or header. | – |
| Chat APIApp service | Load balancer forwards to it for /chat.Calls GPU serving pool.Little's law: throughput ≈ concurrency ÷ latency. 200 slots at 50 ms serve about 4,000 req/s. | 3 replicas × 500 slots · timeout 2.0 s |
| Background job APIApp service | Load balancer forwards to it for /background.Calls Ticket-triage agents.Little's law: throughput ≈ concurrency ÷ latency. 200 slots at 50 ms serve about 4,000 req/s. | 3 replicas × 500 slots · timeout 180 s |
| Ticket-triage agentsAgent orchestrator | Background job API calls it.Calls GPU serving pool.An agent loops: think, call a tool, read the result, repeat until done. | 2000 slots · 8 steps |
| GPU serving poolSelf-hosted serving pool | Chat API calls it. Ticket-triage agents calls it.GPU servers batch requests together; throughput rises with batch size until memory runs out. | 64 slots |