Inference & Serving · core
Continuous Batching
Iteration-level scheduling, dynamic admission, prefill/decode interleaving, chunked prefill, and fairness.
Mental model
Continuous batching changes the batch after each decoding step, filling freed slots without waiting for the slowest request. Throughput improves at the cost of scheduler complexity.
How to study Continuous Batching
Begin by restating the mental model in your own words, then connect it to a concrete system you have built or operated. Name the mechanism, the constraint it addresses, and the trade-off it introduces. Use Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-Serve (OSDI '24), USENIX OSDI '22 — Orca: A Distributed Serving System for Transformer-Based Generative Models, Orca: A Distributed Serving System for Transformer-Based Generative Models (OSDI '22) to check details, but close the source before writing your explanation. Retrieval is the learning step; rereading is only preparation.
Next, compare Continuous Batching with KV Caching & PagedAttention, Model Routing. Ask what changes in correctness, latency, resource use, operability, and failure recovery. Complete Design exercise: Continuous Batching and preserve the command, input, output, and one failed attempt as evidence. Finish by explaining the idea without jargon to someone who has not studied the track.
Proof of understanding
- Explain the mechanism from first principles and identify the state it reads or changes.
- Give one situation where the concept is the right choice and one where it is not.
- Predict a realistic failure mode before running the drill, then compare the prediction with evidence.
- Connect the result to a roadmap or build artifact instead of treating the concept as isolated trivia.
Learn from primary sources
- Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-Serve (OSDI '24) (doc)
- USENIX OSDI '22 — Orca: A Distributed Serving System for Transformer-Based Generative Models (video)
- Orca: A Distributed Serving System for Transformer-Based Generative Models (OSDI '22) (paper)
- SGLang v0.4: Zero-Overhead Batch Scheduler (LMSYS) (article)
- vLLM: Easy, Fast, and Cheap LLM Serving (doc)
Practice and explain it back
Design exercise: Continuous Batching
Iteration-level scheduling, dynamic admission, prefill/decode interleaving, chunked prefill, and fairness. Implement designOutline() returning non-empty values for: admissionPolicy, prefillDecode, fairness. Each value must name a concrete mechanism or decision.
Expected evidence: A design outline with admissionPolicy, prefillDecode, fairness plus an explicit failure mode or trade-off.
Open the interactive drill →Review prompts
- What does continuous batching change relative to static batching, and where does the throughput actually come from?
Build evidence
Synthesize: Inference & Serving
Build a production mental model for inference engines, memory, kernels, routing, hardware, and serving economics. Produce one working system, benchmark, or evidence-backed design that integrates the path.
- Implements or precisely models the core mechanisms from all three milestones
- Includes at least one injected failure or adversarial case and demonstrates recovery
- Reports quality, latency, resource, reliability, or usability measurements relevant to the domain
- Ships a concise architecture note explaining decisions, trade-offs, and remaining risks