How Far Can Disaggregation Go? A Design-Space Exploration of Attention-FFN Disaggregation for Efficient MoE LLM Serving

• URL: http://arxiv.org/abs/2605.28302v1
• Date: Wed, 27 May 2026 10:55:57 GMT
• Title: How Far Can Disaggregation Go? A Design-Space Exploration of Attention-FFN Disaggregation for Efficient MoE LLM Serving
• Abstract summary: We study the benefits and limits of Attention-FFN Disaggregation (AFD) at scale through a framework that fuses on-device kernel measurements with high-fidelity network simulation.<n>Under strict TTFT/TPOT SLOs, AFD sustains around 4k tokens/s of system throughput on DeepSeek-V3.2 across chat, coding, and agentic-coding workloads.
• Score: 14.552785121277529
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: Modern large language model (LLM) inference has progressively disaggregated to keep pace with growing model sizes and tight TTFT and TPOT service-level objectives: from chunked-prefill aggregation, to prefill-decode (P/D) disaggregation, and most recently to operator-level Attention-FFN Disaggregation (AFD). This trend is especially important for mixture-of-experts (MoE) models, where memory-bound attention, compute-intensive expert FFNs, and MoE dispatch/combine communication create distinct resource demands. AFD further exposes this heterogeneity by placing attention and MoE-FFN execution on separate GPU groups. Each level of disaggregation deepens the scheduling design space across workload characteristics, resource allocation, and interconnect topology, raising the central question: when does each level actually pay off? We systematically characterize this trade-off for MoE inference across realistic workloads spanning input/output sequence lengths, prefix-KV reuse, and per-user latency constraints. Using chunked-prefill and P/D disaggregation as baselines, we study the benefits and limits of AFD at scale through a framework that fuses on-device kernel measurements with high-fidelity network simulation. Under strict TTFT/TPOT SLOs, AFD sustains around 4k tokens/s of system throughput on DeepSeek-V3.2 across chat, coding, and agentic-coding workloads, where non-AFD deployments are infeasible. We distill concrete takeaways for jointly optimizing throughput and interactivity, including how to partition attention and FFN across GPUs as a function of workload and model architecture, providing design principles for current rack- and cluster-scale deployments as well as future disaggregated AI infrastructure.

Related papers

• LLM Inference in a Flash! [66.43896490006965]
  Large Language Models (LLMs) have shown impressive capabilities across a range of natural language processing tasks.<n>We present an end-to-end integer-only quantization approach to eliminate expensive floating-point computations.
  arXiv  Detail & Related papers  (2026-09-14T18:04:30Z)
• FLARE: Diffusion for Hybrid Language Model [72.60770374799634]
  FLARE is a systematic conversion framework for hybrid-attention large language models.<n>It combines a token-equal AR-and-diffusion objective, hardware-aware kernels, and unified inference, enabling one checkpoint to support both AR-style verified decoding and diffusion-style parallel denoising.<n>Our results suggest practical dLLMs are limited not only by decoding algorithms, but also by transfer data quality and the training inefficiency of current block-diffusion objectives.
  arXiv  Detail & Related papers  (2026-06-01T06:58:15Z)
• Heterogeneous Tasks Offloading in Vehicular Edge Computing: A Federated Meta Deep Reinforcement Learning Approach [2.7564583149786075]
  Vehicular edge computing (VEC) enables latency-sensitive vehicular applications by offloading computation-intensive tasks to nearby edge servers.<n>In this paper, we propose a Federated Meta Deep Reinforcement Learning framework with GAT-Seq2Seq modeling (FedMAGS) for heterogeneous task offloading in VEC systems.
  arXiv  Detail & Related papers  (2026-05-18T14:08:51Z)
• From AR to Diffusion: Efficiently Adapting Large Language Models with Strictly Causal and Elastic Horizons [74.25767873168745]
  Diffusion models promise efficient parallel text generation but rely on bidirectional attention, creating a structural mismatch with pre-trained Autoregressive (AR) models.<n>We propose FLUID, a framework that efficiently adapts AR backbones to the diffusion paradigm.<n> Experiments demonstrate that FLUID achieves state-of-the-art performance while reducing training costs by orders of magnitude.
  arXiv  Detail & Related papers  (2026-04-11T13:18:27Z)
• Heterogeneous Computing: The Key to Powering the Future of AI Agent Inference [12.811151742829137]
  We introduce two metrics - Operational Intensity (OI) and Capacity Footprint (CF)<n>OI/CF can shift dramatically, with long context KV cache making decode highly memory bound.<n>We further hypothesize agent- hardware co design, multiple inference accelerators within one system, and high bandwidth, large capacity memory disaggregation as foundations for adaptation to evolving OI/CF.
  arXiv  Detail & Related papers  (2026-01-29T17:11:46Z)
• Theoretically Optimal Attention/FFN Ratios in Disaggregated LLM Serving [12.150964112168582]
  We develop a tractable analytical framework for sizing AFD bundles in a $r$A-$1$F topology.<n>We derive closed-form rules for the optimal A/F ratio that maximize average throughput per instance.<n>A trace-calibrated AFD simulator validates the theory.
  arXiv  Detail & Related papers  (2026-01-29T07:22:27Z)
• FLEX: A Backbone for Diffusion-Based Modeling of Spatio-temporal Physical Systems [51.15230303652732]
  FLEX (F Low EXpert) is a backbone architecture for generative modeling of-temporal physical systems.<n>It reduces the variance of the velocity field in the diffusion model, which helps stabilize training.<n>It achieves accurate predictions for super-resolution and forecasting tasks using as few features as two reverse diffusion steps.
  arXiv  Detail & Related papers  (2025-05-23T00:07:59Z)
• FLAMES: A Hybrid Spiking-State Space Model for Adaptive Memory Retention in Event-Based Learning [16.60622265961373]
  FLAMES is a hybrid framework integrating structured state-space dynamics with event-driven computation.<n>By bridging neuromorphic computing and structured sequence modeling, FLAMES enables scalable long-range reasoning in event-driven systems.
  arXiv  Detail & Related papers  (2025-04-02T00:08:19Z)
• Let Synthetic Data Shine: Domain Reassembly and Soft-Fusion for Single Domain Generalization [68.41367635546183]
  Single Domain Generalization aims to train models with consistent performance across diverse scenarios using data from a single source.<n>We propose Discriminative Domain Reassembly and Soft-Fusion (DRSF), a training framework leveraging synthetic data to improve model generalization.
  arXiv  Detail & Related papers  (2025-03-17T18:08:03Z)
• Task-Oriented Feature Compression for Multimodal Understanding via Device-Edge Co-Inference [54.53508601749513]
  We propose a task-oriented feature compression (TOFC) method for multimodal understanding in a device-edge co-inference framework.<n>To enhance compression efficiency, multiple entropy models are adaptively selected based on the characteristics of the visual features.<n>Results show that TOFC achieves up to 52% reduction in data transmission overhead and 63% reduction in system latency.
  arXiv  Detail & Related papers  (2025-03-17T08:37:22Z)
• FusionLLM: A Decentralized LLM Training System on Geo-distributed GPUs with Adaptive Compression [55.992528247880685]
  Decentralized training faces significant challenges regarding system design and efficiency. We present FusionLLM, a decentralized training system designed and implemented for training large deep neural networks (DNNs) We show that our system and method can achieve 1.45 - 9.39x speedup compared to baseline methods while ensuring convergence.
  arXiv  Detail & Related papers  (2024-10-16T16:13:19Z)
• Resource Management for Low-latency Cooperative Fine-tuning of Foundation Models at the Network Edge [35.40849522296486]
  Large-scale foundation models (FoMos) can perform human-like intelligence. FoMos need to be adapted to specialized downstream tasks through fine-tuning techniques. We advocate multi-device cooperation within the device-edge cooperative fine-tuning paradigm.
  arXiv  Detail & Related papers  (2024-07-13T12:47:14Z)
• SHERL: Synthesizing High Accuracy and Efficient Memory for Resource-Limited Transfer Learning [63.93193829913252]
  We propose an innovative METL strategy called SHERL for resource-limited scenarios. In the early route, intermediate outputs are consolidated via an anti-redundancy operation. In the late route, utilizing minimal late pre-trained layers could alleviate the peak demand on memory overhead.
  arXiv  Detail & Related papers  (2024-07-10T10:22:35Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.