PUZZLE: Efficiently Aligning Large Language Models through Light-Weight Context Switching
Efficient system for aligning large language models, optimizing model orchestration and workload switching to minimize overhead and improve performance in AI alignment tasks.
Fast Inference for Probabilistic Graphical Models
Innovative system for efficient probabilistic graphical model inference, offering significant speedup through optimized memory management, parallelization, and flexible integration with sampling algorithms.
SimEnc - High-Performance Similarity-Preserving Encryption for Encrypted Docker Image Deduplication
Innovative encryption technique for Docker image deduplication, enhancing storage efficiency and pull performance while preserving privacy in registries.
FlexMem - Adaptive Page Profiling and Migration for Tiered Memory
Adaptive page profiling and migration system for tiered memory, improving performance through dynamic decision-making and combining profiling methods for enhanced memory management.
ETC: An Elastic Transmission Control Using End-to-End Available Bandwidth Perception
Innovative transport protocol ETC uses pulling rate as congestion signal, outperforming existing protocols in throughput, latency, convergence speed, and fairness for improved network performance.
Scalable and Effective Page-table and TLB Management on NUMA Systems
Innovative approach to page-table management in NUMA systems, enhancing performance through on-demand replication and reduced TLB shootdowns, benefiting highly threaded applications.
MAGPY: Compiling Eager Mode DNN Programs by Monitoring Execution States
Optimizing deep learning programs by generating complete operator graphs through runtime monitoring, enhancing performance for complex Python-based DNN implementations.
FetchBPF - Customizable Prefetching Policies in Linux with eBPF
Explore FetchBPF, a framework for customizing Linux memory prefetching policies using eBPF, enabling performance optimization without kernel modifications.
A Secure, Fast, and Resource-Efficient Serverless Platform with Function REWIND
Innovative serverless platform REWIND enhances security and efficiency by resetting containers, using kernel-level snapshots, and optimizing memory usage, improving performance while maintaining data isolation between function requests.
Fast Trapless Kernel Probes Everywhere - USENIX ATC '24
Efficient trapless kernel probing mechanism for Linux, improving performance 10x over standard Kprobe while covering 96% of kernel code. Enhances debugging, security, and monitoring capabilities.
Conspirator - SmartNIC-Aided Control Plane for Distributed ML Workloads
Innovative control plane design using SmartNICs to optimize distributed ML workloads, reducing bottlenecks and improving resource utilization for enhanced performance and efficiency.
Evaluating Chiplet-based Large-Scale Interconnection Networks via Cycle-Accurate Packet-Parallel Simulation
Explore cycle-accurate simulation of large-scale chiplet networks using CNSim, a novel packet-parallel simulator. Gain insights into heterogeneous network design and evaluation for advanced chiplet architectures.
Pecan: Cost-Efficient ML Data Preprocessing with Automatic Transformation Ordering and Hybrid Placement
Optimize ML data preprocessing costs with Pecan's dynamic worker scheduling and transformation reordering, reducing preprocessing expenses by 87% and total training costs by up to 60%.
Full Lifecycle Data Analysis on a Large-scale Leadership Supercomputer
Comprehensive analysis of Sunway TaihuLight supercomputer data reveals insights into HPC operational management and I/O workload characteristics, offering valuable lessons for system optimization.
KEPC-Push: A Knowledge-Enhanced Proactive Content Push Strategy for Edge-Assisted Video Feed Streaming
Innovative strategy for optimizing video feed streaming using knowledge graphs to predict content popularity and improve resource allocation, resulting in reduced bandwidth costs and faster download speeds.
StreamCache - Revisiting Page Cache for File Scanning on Fast Storage Devices
StreamCache: A new page cache system for fast storage devices, improving I/O performance through stream tracking, efficient reclaiming, and accelerated page allocation. Enhances scientific apps and large language model operations.
PeRF - Preemption-enabled RDMA Framework for Multi-tenant Environments
Innovative RDMA framework for multi-tenant cloud environments, offering performance isolation and work conservation without compromising bare-metal performance through a novel RNIC preemption mechanism.
Kivi: Verification for Cluster Management in Kubernetes
Kivi: A pioneering system for verifying Kubernetes controllers and configurations, addressing challenges in cloud infrastructure management through model checking and optimization techniques.
Empirical Study of Rust-for-Linux: Success, Dissatisfaction, and Compromise
Empirical study on Rust-for-Linux project, analyzing its impact on kernel safety and performance. Explores successes, challenges, and compromises in integrating Rust into the Linux ecosystem.
Harmonizing Efficiency and Practicability: Optimizing Resource Utilization in Serverless Computing
Optimizing serverless computing resource utilization with Jiagu: novel techniques for accurate prediction, efficient scaling, and improved deployment density while maintaining QoS.
Balancing Analysis Time and Bug Detection: Daily Development-friendly Bug Detection in Linux
Explore FiTx, a framework for fast, daily bug detection in Linux. Learn how it balances analysis time and bug-finding capabilities, outperforming existing tools in speed and effectiveness.