#concurrency

Single-user latency was 200ms. At 100 concurrent users, it's 3 seconds. The model didn't slow down. Your serving architecture did.