NCCL-BOT: Closing the Loop on NCCL's Cost Model with Online Bandits
cs.DC, cs.LG
Submitted: 2026-11-08
Updated: 2026-11-08
Comments: 46 pages, 13 figures, 38 tables
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- CCL-Bench 1.0: A Trace-Based Benchmark for LLM Infrastructure
- Demystifying NCCL: An In-depth Analysis of GPU Communication Protocols and Algorithms
- MSCCL++: Rethinking GPU Communication Abstractions for AI Inference
- Algorithms for multi-armed bandit problems
- An efficient algorithm for learning with semi-bandit feedback
- Collective Communication for 100k+ GPUs
- Introduction to Multi-Armed Bandits
- Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC
- ForestColl: Throughput-Optimal Collective Communications on Heterogeneous Network Fabrics
Related papers
- Programmable Probabilistic Computer with 1,000,000 p-bits
- BatchGen: An Architecture for Scalable and Efficient Batch Inference
- Seamless Reconfiguration for DAG-BFT (Technical Report)
- ELDR: Expert-Locality-Aware Decode Routing for PD-Disaggregated MoE Serving
- iScheduler: Reinforcement Learning-Driven Continual Optimization for Large-Scale Resource Investment Problems
- TIRx: A Unified Tile Primitive ML Compiler Across Heterogeneous Backends