131 papers
This paper introduces Structured Gossip DNS, a partition-resilient name resolution system for large-scale dynamic networks that leverages DHT finger tables and passive stabilization to achieve eventual consistency with reduced message complexity and without requiring global coordination.
This paper evaluates the open-source CODECO toolkit, demonstrating that it significantly reduces manual intervention and maintains competitive performance compared to baseline Kubernetes workflows for automating multi-service deployments across heterogeneous Edge-Cloud environments.
This paper proposes a dual spatial-scale UAV network optimization framework that combines exact potential game algorithms for link configuration and deployment with a large language model to dynamically generate utility weights, thereby enhancing adaptability and performance in terms of energy efficiency, latency, and throughput.
This paper argues that the current fragmentation in high-performance interconnects stems from a fundamental "semantic crisis" caused by implicit, vendor-specific assumptions about time and ordering, and proposes that adopting explicit, testable link semantics through the Open Atomic Ethernet (OAE) standard is essential to restore correctness and enable convergence.
Uber's Failover Architecture (UFA) replaces its costly uniform 2x capacity model with a differentiated, criticality-based approach that opportunistically shares resources and preempts non-critical services during peak failovers, thereby reducing steady-state provisioning from 2x to 1.3x and eliminating over one million CPU cores while maintaining 99.97% availability.
The paper introduces AIReSim, a discrete event simulator designed to help system designers evaluate and tune reliability mechanisms, prioritize improvements, and plan capacity for large-scale AI clusters by modeling the complex tradeoffs involved in failure, recovery, scheduling, and repair processes.
This paper introduces a configuration-driven runtime orchestration framework that dynamically generates execution graphs from configuration at request time, enabling low-latency, dependency-aware parallel data retrieval across distributed microservices and APIs without requiring workflow code redeployment.
This paper presents a fully device-resident, multi-GPU architecture that achieves near-zero host-device communication and 99.7% parallel efficiency through lock-free work-stealing and optimized shared-memory tiling, enabling the verification of Goldbach's conjecture up to $10^{13}$ in just 133.5 seconds on a four-GPU system with a 45.6× speedup over previous methods.
The paper presents NANOMIND, a hardware-software co-design framework that decomposes Large Multimodal Models into modular components and dynamically schedules them across heterogeneous accelerators on unified-memory SoCs, enabling a battery-powered device to run LMMs entirely on-device with significantly improved energy efficiency and throughput.
ArcLight is a lightweight LLM inference architecture designed specifically for many-core CPUs that overcomes cross-NUMA memory access bottlenecks through efficient memory management, thread scheduling, and controlled tensor parallelism, achieving up to 46% higher throughput than mainstream frameworks while maintaining broad device compatibility.
This paper presents EROICA, the first online troubleshooting system deployed on production-scale GPU clusters (~100,000 GPUs) that effectively diagnoses complex hardware and software performance issues in large-scale model training through fine-grained profiling and differential observability with minimal impact.
This paper theoretically demonstrates that in distributed computing systems with probabilistic stragglers, the approximation errors of Berrut Approximate Coded Computing (BACC) and Learning Theoretic Coded Computing (LeTCC) schemes converge to zero at specific rates despite the average number of stragglers scaling with the total server count, a finding validated through experiments on various functions including deep neural networks.
This paper introduces Co-LoRA, a collaborative personalization framework that addresses both data and model heterogeneity through a task-relevance-aware aggregation strategy and a dimension-invariant module, validated by a new multi-modal benchmark and superior performance over state-of-the-art methods.
The paper introduces Covenant-72B, a 72-billion-parameter language model successfully pre-trained on 1.1 trillion tokens through the largest permissionless, globally distributed collaboration to date, demonstrating that open, blockchain-supported participation can achieve performance competitive with centralized training at unprecedented scale.
This paper presents Megatron Core, a scalable and production-ready open-source framework that addresses the coupled memory, communication, and computation challenges of Mixture-of-Experts (MoE) training through integrated system-level optimizations, enabling high-performance training of models ranging from billions to trillions of parameters on large-scale GPU clusters.
This dissertation addresses the memory bottleneck in modern computing by advocating a shift from data-agnostic to data-informed microarchitectural designs, proposing four machine learning-driven and data-aware mechanisms that significantly enhance performance and energy efficiency.
This paper introduces MAS-H2, a hierarchical multi-agent system for Kubernetes that bridges the gap between business policies and resource provisioning through strategic, planning, and execution agents, demonstrating significant reductions in CPU stress and peak load while enabling zero-downtime infrastructure migrations compared to native autoscalers.
The paper proposes TA-RNN-Medical-Hybrid, a time-aware and interpretable deep learning framework that integrates continuous-time encoding, SNOMED-based disease representations, and a hierarchical dual-level attention mechanism to accurately predict ICU mortality risk while providing clinically meaningful explanations.
NEST is a novel device placement framework that unifies network, compute, and memory awareness through structured dynamic programming to jointly optimize hybrid parallelism strategies, achieving up to 2.43 times higher throughput and improved scalability compared to state-of-the-art baselines.
This paper argues that resilience must be established as a fundamental design constraint in integrated classical-quantum computing systems by developing new quantitative models and metrics, drawing inspiration from civil engineering to accurately assess the cost-benefit ratios of system improvements and their cascading impacts on end-user value.