Zhong et al.: SLO-aware prefill/decoding disaggregation

What changed

Placed prefill and decoding on different GPUs and jointly selected per-phase resource allocation, parallelism and placement to improve goodput subject to first-token and per-output-token latency targets.

Scientific scope

Placed prefill and decoding on different GPUs and jointly selected per-phase resource allocation, parallelism and placement to improve goodput subject to first-token and per-output-token latency targets. Disaggregation moves KV states and incurs communication/extra-resource costs. Original v1 metrics differ from later versions; no exact speedup exported. Goodput under selected SLOs is distinct from raw throughput or per-request latency.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Disaggregation moves KV states and incurs communication/extra-resource costs. Original v1 metrics differ from later versions; no exact speedup exported. Goodput under selected SLOs is distinct from raw throughput or per-request latency.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Serving

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction