REVIEW 19 cited by
MegaScale-Infer: Serving Mixture-of-Experts at Scale with Disaggregated Expert Parallelism
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
MegaScale-Infer: Serving Mixture-of-Experts at Scale with Disaggregated Expert Parallelism
read the original abstract
Mixture-of-Experts (MoE) showcases tremendous potential to scale large language models (LLMs) with enhanced performance and reduced computational complexity. However, its sparsely activated architecture shifts feed-forward networks (FFNs) from being compute-intensive to memory-intensive during inference, leading to substantially lower GPU utilization and increased operational costs. We present MegaScale-Infer, an efficient and cost-effective system for serving large-scale MoE models. MegaScale-Infer disaggregates attention and FFN modules within each model layer, enabling independent scaling, tailored parallelism strategies, and heterogeneous deployment for both modules. To fully exploit disaggregation in the presence of MoE's sparsity, MegaScale-Infer introduces ping-pong pipeline parallelism, which partitions a request batch into micro-batches and shuttles them between attention and FFNs for inference. Combined with distinct model parallelism for each module, MegaScale-Infer effectively hides communication overhead and maximizes GPU utilization. To adapt to disaggregated attention and FFN modules and minimize data transmission overhead (e.g., token dispatch), MegaScale-Infer provides a high-performance M2N communication library that eliminates unnecessary GPU-to-CPU data copies, group initialization overhead, and GPU synchronization. Experimental results indicate that MegaScale-Infer achieves up to 1.90x higher per-GPU throughput than state-of-the-art solutions.
Forward citations
Cited by 19 Pith papers
-
Sieve: Dynamic Expert-Aware PIM Acceleration for Evolving Mixture-of-Experts Models
Sieve dynamically schedules MoE experts across GPU and PIM hardware to handle bimodal token distributions, achieving 1.3x to 1.6x gains in throughput and interactivity over static prior PIM systems on three large models.
-
SmoothAgent: Efficient Long-Horizon LLM-Based Agent Serving with Lookahead Context Engineering
SmoothAgent introduces lookahead context engineering to eliminate transformation overhead in LLM agents, reducing TTFT by up to 11.9x through proactive KV cache preparation.
-
ViBE: Co-Optimizing Workload Skew and Hardware Variability for MoE Serving
ViBE co-optimizes expert placement with measured GPU performance variability in MoE inference to cut execution-time imbalance, delivering 14% better SLO attainment and up to 45% lower P90 TTFT.
-
Frontier: Towards Comprehensive and Accurate LLM Inference Simulation
Frontier is a new discrete-event simulator for disaggregated LLM serving that incorporates co-location, PDD, AFD, and optimizations, achieving under 4% throughput error and large reductions in latency prediction error...
-
Surviving Partial Rank Failures in Wide Expert-Parallel MoE Inference
EEP makes wide expert-parallel MoE serving survive single-rank failures with an 11s recovery pause, 8s reintegration pause, and throughput restored to 95% of pre-fault level within 52s while staying within 4.4% of a f...
-
Analytical Provisioning for Attention-FFN Disaggregated LLM Serving under Stochastic Workloads
A renewal-reward analysis yields a closed-form mean-field rule for the optimal Attention/FFN provisioning ratio in disaggregated LLM serving that accounts for stochastic KV-cache growth and matches simulation optima w...
-
UBEP: Re-architecting Expert Parallelism Communication Library for Production Superpods
UBEP replaces BSP All-to-All for MoE on multi-tier superpods with dependency-driven kernel decomposition, topology-aware token scheduling, and Data-as-Flag atomics, cutting All-to-All latency up to 52.4% and TPOT up to 11.1%.
-
UBEP: Re-architecting Expert Parallelism Communication Library for Production Superpods
UBEP re-architects MoE All-to-All communication for superpods via kernel decomposition, topology-aware scheduling, and data-as-flag synchronization, achieving up to 52.4% latency reduction on CM384 hardware.
-
Think Before You Grid-Search: Floor-First Triage for LLM Serving
A five-dimensional resource-vector floor model computes latency bounds and capacity walls for LLM serving, predicting when TP16 or EP16+DP attention layouts dominate based on operating point.
-
Think Before You Grid-Search: Floor-First Triage for LLM Serving
LLM serving should triage by five-resource analytical floors and wall ordering, not grid search; on 16×H20, TP16 is capacity-capped at ~70 while EP+DP attention reaches ~644 concurrent 8K requests.
-
DisagMoE: Computation-Communication overlapped MoE Training via Disaggregated AF-Pipe Parallelism
DisagMoE achieves up to 1.8x faster MoE training by disaggregating attention and FFN layers into disjoint GPU groups with a multi-stage uni-directional pipeline and roofline-based bandwidth balancing.
-
AMMA: A Multi-Chiplet Memory-Centric Architecture for Low-Latency 1M Context Attention Serving
AMMA is a memory-centric multi-chiplet architecture using HBM-PNM cubes, custom logic dies, hybrid parallelism, and reordered collectives that delivers 15.5X lower attention latency and 6.9X lower energy than NVIDIA H...
-
CascadeInfer: Length-Aware Scheduling of LLM Serving with Low Latency and Load Balancing
CascadeInfer partitions LLM instances into length-specialized groups, uses dynamic programming for stage partitioning, and applies runtime refinement plus decentralized load balancing to cut latency and raise throughput.
-
Patterns behind Chaos: Forecasting Data Movement for Efficient Large-Scale MoE LLM Inference
Comprehensive profiling of expert selection in frontier MoE models reveals temporal and spatial patterns that enable 6.6x speedup on wafer-scale GPUs and 1.25x on existing systems via targeted optimizations.
-
KernelFlume: Elastic Core-Attention Scaling for Agentic Long-Context Decoding
KernelFlume presents a disaggregated decode architecture that separates core attention from projection/FFN paths to enable elastic scaling of attention nodes, reporting up to 61% lower cost per million tokens versus f...
-
How Far Can Disaggregation Go? A Design-Space Exploration of Attention-FFN Disaggregation for Efficient MoE LLM Serving
Operator-level attention-FFN disaggregation enables ~4k tokens/s throughput for DeepSeek-V3.2 under tight TTFT/TPOT SLOs where chunked-prefill and prefill-decode baselines cannot.
-
UniEP: Unified Expert-Parallel MoE MegaKernel for LLM Training
UniEP fuses MoE communication and computation into unified MegaKernels with deterministic token ordering, delivering 1.03x-1.38x speedups over prior work while preserving training accuracy.
-
Understanding and Improving Communication Performance in Multi-node LLM Inference
Performance analysis of multi-node LLM inference identifies all-reduce bottlenecks and introduces NVRAR hierarchical all-reduce achieving 1.9-3.6x lower latency than NCCL and up to 1.72x end-to-end batch latency reduc...
-
Position: LLM Serving Needs Mathematical Optimization and Algorithmic Foundations, Not Just Heuristics
LLM serving requires mathematical optimization and algorithms with provable guarantees rather than generic heuristics that fail unpredictably on LLM workloads.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.