Research publications
From the literature
Research publications
Deep dives are papers that use, extend, or benchmark Dynamo. Citations are papers that cite it in their related work without building on it.
PACT '26University of GeorgiaOct 2026EAServe: Encode-Aware Disaggregated Serving for Multimodal Large Language ModelsarXivNVIDIASep 21, 2026Fast Recovery for LLM Serving via Decoupled Device Memory Lifetime in DynamoarXivBoston UniversitySep 19, 2026From Inference Engine to Inference Control Plane: Connecting vLLM, llm-d, and the Evolution of Efficient Distributed LLM ServingarXivSep 14, 2026Calibrate, Then Route: A Measured Study of Learned Request Routing for Disaggregated LLM ServingarXivXenoscubeSep 10, 2026Phase-Decoupled, Model-Calibrated Power Control for Disaggregated LLM ServingSIGCOMM '26 NIACGeorgia TechAug 2026ARK: Avoiding Routing Collisions for KV Cache Transfer in Disaggregated LLM InferencearXivMarvellJul 29, 2026A Photonic-CXL Memory Appliance for Scalable KV Cache Management in LLM InferencearXivSK hynixJul 20, 2026A CXL Memory Rack for Multi-Turn LLM ServingOSDI '26Korea UniversityJul 13, 2026Revisiting Pipeline Parallelism for LLM ServingOSDI '26Shanghai Jiao Tong UniversityJul 13, 2026Simple is Better: Multiplication May Be All You Need for LLM Request SchedulingHPDC '26Nanjing UniversityJul 13, 2026STAR: Decode-Phase Rescheduling for LLM InferencearXivShanghai Jiao Tong UniversityJul 9, 2026SMetric: Rethink LLM Scheduling for Serving Agents with Balanced Session-centric SchedulingarXivEmerald AIJun 23, 2026Power-Flexible AI Data Centers: A New Paradigm for Grid-Responsive ComputearXivSolyx AIJun 13, 2026Solyx AI Grid: Hardware-Telemetry-Aware Routing Across Geographically Distributed GPU ClustersarXivJun 11, 2026The Price of Anarchy in Disaggregated InferencearXivNational College of IrelandJun 2, 2026NetKV: Network-Aware Decode Instance Selection for Disaggregated LLM InferenceMLSys 2026May 2026Breaking the Ice: Analyzing Cold Start Latency in vLLMMLSys 2026NVIDIAMay 2026A Pragmatic Exploration of Prefill-Decode Disaggregation in Large Scale InferenceMLSys 2026GroqMay 2026SHIP: SRAM-Based Huge Inference Pipelines for Fast LLM ServingMLSys 2026Perplexity AIMay 2026fabric-lib: RDMA Point-to-Point Communication for LLM SystemsarXivMay 16, 2026ObjectCache: Layerwise Object-Storage Retrieval for KV Cache ReuseFrontiers of Computer ScienceLenovoMay 9, 2026Adaptive Parallelism for LLM Inference with Model Irrelevant ProfilerEuroMLSys '26IBM ResearchApr 27, 2026A Case for a Simulation-Driven Exploration of Distributed GenAI PlatformsarXivMPI-SWSApr 24, 2026LaissezCloud: Continuous Resource Renegotiation for the Public CloudarXivApr 16, 2026Prefill-as-a-Service: KVCache of Next-Generation Models Could Go Cross-DatacenterarXivApr 1, 2026TENT: A Declarative Slice Spraying Engine for Performant and Resilient Data Movement in Disaggregated LLM ServingarXivMar 13, 2026NCCL EP: Towards a Unified Expert Parallel Communication API for NCCLIEEE TSCShenzhen Institutes of Advanced Technology, CASMar 2026DOPD: A Dynamic PD-Disaggregation Architecture for Maximizing Goodput in LLM Inference ServingICML 2026Feb 16, 2026Efficient Multi-round LLM Inference over Disaggregated ServingarXivNVIDIAFeb 14, 2026ThunderAgent: A Simple, Fast and Program-Aware Agentic Inference SystemICML 2026University of CambridgeFeb 12, 2026OServe: Accelerating LLM Serving via Spatial-Temporal Workload OrchestrationarXivNVIDIAJan 9, 2026AIConfigurator: Lightning-Fast Configuration Optimization for Multi-Framework LLM ServingIEEE AccessAmazon2026Optimizing GPU Workloads on Kubernetes: An Integrated Approach Using NVIDIA Dynamo, run:ai (KAI), and Amazon EKSarXivSK HynixDec 20, 2025TraCT: Disaggregated LLM Serving with CXL Shared Memory KV Cache at Rack-ScalearXivNov 6, 2025DuetServe: Harmonizing Prefill and Decode for LLM Serving via Adaptive GPU MultiplexingarXivCelestica AIOct 31, 2025AMD MI300X GPU Performance AnalysisarXivCapital OneOct 16, 2025From Attention to Disaggregation: Tracing the Evolution of LLM InferencearXivOct 15, 2025BanaServe: Unified KV Cache and Dynamic Module Migration for Balancing Disaggregated LLM Serving in AI InfrastructurearXivLMCacheOct 8, 2025LMCache: An Efficient KV Cache Layer for Enterprise-Scale LLM InferenceIEEE eScience 2025Illinois Institute of TechnologySep 15, 2025Insights into GPUDirect Data Transfer through NIXL BenchmarkingarXivEPFLAug 22, 2025GreenLLM: SLO-Aware Dynamic Frequency Scaling for Energy-Efficient LLM ServingIEEE MicroMay 2025Toward Disaggregated and Heterogenous AI SystemsarXivInfinigence-AIApr 28, 2025semi-PD: Towards Efficient LLM Serving via Phase-Wise Disaggregated Computation and Unified Storage