TEMPO: Makespan-Aware Expert-Parallel Load Balancing Across Memory- and Compute-Bound Regimes
On the map: Mixture-of-experts → Efficient MoE training and inference systems
Abstract (arXiv)
In expert-parallel (EP) MoE serving, every layer synchronizes at the slowest GPU. Dispatchers balance token counts (EPLB, LPLB, UltraEP) or activated-expert counts (METRO), assuming expert time is linear in one. Measurements on two datacenter GPU generations show it is neither: below n^* ≈ 156--168 tokens, HBM weight streaming dominates---cost attaches to activated replicas, not tokens; above it, grouped GEMM rounds tokens to 128-tile M-tiles, so splitting an expert adds padded compute. A max-affine profile t=max(a+bG,,c+β N) captures both regimes. Realistic decode batches hold hot experts in the linear regime and cold in the flat simultaneously; recorded batches show proxy dispatches differ by 1.4--1.6× in modeled block time (p95 up to 1.7×), and which proxy wins flips with the regime. We formalize per-batch dispatch as a fixed-charge makespan problem---NP-hard on two fully replicated GPUs, polynomial in degenerate limits---and present TEMPO, a makespan-aware dispatcher solving it in milliseconds off the critical path; its SGLang integration runs out-of-process and fuses dispatch with count collection into one in-graph kernel. Anchored by an 8-GPU Testbed A microbenchmark, TEMPO stays within 1% of the best fixed baseline everywhere and wins by up to 15.5% where regimes mix. End-to-end on Testbed B, Qwen3-235B (inside the win region) gains 4--6% throughput and cuts p99 latency by ~ 15.6%; DeepSeek-V3 (outside, communication-dominated) shows only mechanism cost. A phase diagram, not a universal win, is the claim: it predicts both outcomes before deployment.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Efficient MoE training and inference systems, one of 12 directions in Mixture-of-experts. Builds on 12 of its 22 references that are in the map. First posted on arXiv 13 Aug 2026.
Builds on
12 of 22 references.
Closest ancestors · 7
Same direction, newest first
- 2026TAOT: Topology-Aware Optimal Transport for Dynamic Expert Replica Placement in MoE Training4 Aug
- 2026Beyond Uniform Experts: Cost-Aware Expert Execution for Efficient Multi-Device MoE Inference29 Jun
- 2026UltraEP: Unleash MoE Training and Inference on Rack-Scale Nodes with Near-Optimal Load Balancing2 Jun
- 2026ViBE: Co-Optimizing Workload Skew and Hardware Variability for MoE Serving30 May
- 2026Least-Loaded Expert Parallelism: Load Balancing An Imbalanced Mixture-of-Experts23 Jan
- 2025Efficient MoE Serving in the Memory-Bound Regime: Balance Activated Experts, Not Tokens10 Dec
- 2025Taming Latency-Memory Trade-Off in MoE-Based LLM Serving via Fine-Grained Expert Offloading7 Feb
Other references · 5
- 2017Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layercited by 3,509
- 2021Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsitycited by 3,073
- 2020GShard: Scaling Giant Models with Conditional Computation and Automatic Shardingcited by 1,631
- 2023FlexMoE: Scaling Large-scale Sparse Pre-trained Model Training via Dynamic Device Placementcited by 53
- 2026Mixture-of-Experts Servingcited by 2
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2026FEPLB: Exploiting Copy Engines for Nearly Free MoE Load Balancing in Distributed Training
- 2026TopoEP: Topology-Aware Load Balancing for Expert-Parallel MoE Training
- 2026Diagnosing Overhead in Dispatch Operations: Cross-architecture Observatory
- 2025Toward Cost-Efficient Serving of Mixture-of-Experts with Asynchrony
- 2025MoETuner: Optimized Mixture of Expert Serving with Balanced Expert Placement and Token Routing
- 2026Flattening Every Memory Peak in Long-Context Mixture-of-Experts Training
- 2025Advancing MoE Efficiency: A Collaboration-Constrained Routing (C2R) Strategy for Better Expert Parallelism Design
- 2026WiSP: A Working-Set View of Mixture-of-Experts Serving on Extremely Low-Resource Hardware