Llm Inference Serving easy
You are adding streaming to a text-completion API. The client sends one prompt, and the server streams tokens back through your existing HTTP load balancers and proxies. Which transport fits this pattern best?
- A WebSocket, because a full-duplex connection is required for the server to push tokens as they are generated.
- B Server-Sent Events, because one-way server-to-client streaming over plain HTTP fits a single request-response.
- C Long polling, because each poll returns the tokens produced since the previous one without holding connections.
- D gRPC bidirectional streaming, because browsers natively support it over HTTP/1.1 for incremental token delivery.
Answer: B. Server-Sent Events, because one-way server-to-client streaming over plain HTTP fits a single request-response.
**SSE** streams server-to-client over ordinary HTTP, so it works with existing proxies, auth, and load balancers with little extra machinery. WebSocket is tempting, but full-duplex is not required for push, and it adds connection-upgrade and stateful-routing complexity. WebSocket is the right choice when the client must send data mid-stream, such as live audio or interruptions. Long polling adds a round trip per batch of tokens, and browsers cannot do gRPC bidirectional streaming natively.
One system design question a day, with spaced repetition so the ones you miss come back.
Get the appRelated questions
- Your 70B-class model servers each handle about 40 concurrent chats with 4K-token contexts. Product raises the maximum context to 32K tokens, and many users start using it. What is the main capacity consequence?
- A 70B-class model (80 layers, grouped-query attention with 8 KV heads of dimension 128) is served in fp16 on one 8-GPU node with about 640 GB of total HBM. Product wants 256 concurrent sessions, each holding a full 32K-token context. What is the binding constraint, and what does it imply?
- Your server reserves a contiguous KV region sized for max_tokens = 8,192 for every request at admission, but the average sequence ends near 600 tokens. The node reports out-of-KV-memory at about 20 concurrent requests while GPU compute sits at 30%. Which change best raises concurrency without restricting legitimate long outputs?
- You must summarize 10 million documents overnight, and no user is waiting on any single result. How should you configure the serving stack compared with your interactive chat deployment?
- You serve a 7B-parameter model in fp16 (about 14 GB of weights) for a chat app. Users say single-conversation token generation is too slow, and you are choosing new GPU hardware. Which hardware spec most directly increases per-sequence decode speed?
- Your RAG workload sends 8K-token prompts and gets ~150-token answers from a 70B-class model. You plan to move prefill and decode onto separate GPU pools so long prefills stop stalling decode. What new resource must you size carefully?