HydraRPC - Efficient RPC System Using CXL-Attached HDM
Explores HydraRPC, utilizing CXL-attached HDM for efficient data transmission, eliminating network rounds and memory copying. Evaluates performance on real CXL hardware, showcasing potential for building efficient RPC systems.
SlimArchive - A Lightweight Architecture for Ethereum Archive Nodes
Innovative approach to optimize Ethereum archive nodes, reducing storage by 98.1% and improving performance through compacted, flattened, and fine-grained historical state maintenance.
UniMem - Redesigning Disaggregated Memory within a Unified Local-Remote Memory Hierarchy
Explores UniMem, a novel disaggregated memory system that optimizes remote memory access, local cache utilization, and page migration for improved performance in large-scale computing environments.
Taming Hot Bloat Under Virtualization with HUGESCOPE
Innovative solution for addressing hot bloat in virtualization, improving performance and memory efficiency. HugeScope offers lightweight, effective framework for page tracking, splitting, and coalescing on commodity hardware.
Quant-LLM: Accelerating Large Language Model Serving via FP6-Centric Algorithm-System Co-Design
Innovative GPU kernel design enables efficient 6-bit quantization for large language models, significantly improving inference speed while maintaining quality across various applications.
ALPS: An Adaptive Learning, Priority OS Scheduler for Serverless Functions
Innovative OS scheduler for serverless functions, combining SRPT and proportional-share scheduling to optimize performance of short-lived tasks while preventing starvation of longer functions.
FBMM: Making Memory Management Extensible With Filesystems
Innovative approach using filesystems for extensible memory management, enabling diverse memory configurations and policies with minimal code and competitive performance.
Metis - Fast Automatic Distributed Training on Heterogeneous GPUs
Efficient distributed training system for heterogeneous GPUs. Optimizes parallelism plans, improves resource utilization, and achieves significant speed-ups for large deep learning models.
WingFuzz: Implementing Continuous Fuzzing for Database Management Systems
Continuous fuzzing technique for enterprise-level DBMSs, addressing challenges in SQL grammar, codebase evolution, and anomaly analysis. Discovered 236 bugs across 12 major DBMSs.
Data Caching for Enterprise-Grade Petabyte-Scale OLAP
Explores Alluxio local cache for petabyte-scale OLAP, optimizing data transfer efficiency in decoupled compute-storage models. Shares insights from Uber and Meta deployments.
Ethane - An Asymmetric File System for Disaggregated Persistent Memory
Innovative file system leveraging disaggregated persistent memory to enhance performance and cost-effectiveness in distributed environments, outperforming modern alternatives by up to 68x.
MSFRD: Mutation Similarity Based SSD Failure Rating and Diagnosis for Complex Production Environments
Innovative SSD failure prediction scheme using mutation similarity for data centers. Improves precision and recall, enables early rating and progressive diagnosis for enhanced storage reliability.
CPC: Flexible, Secure, and Efficient CVM Maintenance with Confidential Procedure Calls
Innovative approach for secure and efficient maintenance of Confidential Virtual Machines, introducing Confidential Procedure Calls to enhance performance, security, and flexibility in cloud computing environments.
TileClipper: Lightweight Selection of Regions of Interest from Videos for Traffic Surveillance
Innovative video surveillance system reducing data sent to cloud by 22% while maintaining 92% detection accuracy. Utilizes tile sampling for efficient traffic monitoring on low-cost edge devices.
Enhancing Resource Management of the World's Largest PCDN System for On-Demand Video Streaming
Explore ByteDance's PCDN+ system, enhancing video content delivery with cost-efficient edge resources. Learn about its centralized/decentralized framework, improved resource utilization, and higher service speeds.
Accelerating the Training of Large Language Models - Efficient Activation Rematerialization and Optimal Hybrid Parallelism
Innovative strategies for accelerating large language model training: efficient activation rematerialization and optimal hybrid parallelism, boosting performance and resource utilization.
Config-Snob: Tuning for the Best Configurations of Networking Protocol Stack
Optimizing web server performance through dynamic tuning of networking protocol stack configurations, addressing challenges in configuration space and network dynamics.
Expeditious High-Concurrency MicroVM SnapStart in Persistent Memory with Augmented Hypervisor
Innovative system PASS leverages persistent memory for efficient MicroVM SnapStart execution, reducing execution time and doubling concurrency compared to existing solutions.
More is Different: Prototyping and Analyzing a New Form of Edge Server with Massive Mobile SoCs
Innovative edge server design using mobile SoCs for improved energy efficiency in cloud gaming, video transcoding, and deep learning inference, with insights on performance and cost-effectiveness.
CrossMapping - Harmonizing Memory Consistency in Cross-ISA Binary Translation
Innovative framework for cross-ISA binary translation, enhancing emulation correctness and performance. Introduces specification tables for unified memory model description and efficient primitive mapping.
gVulkan - Scalable GPU Pooling for Pixel-Grained Rendering in Ray Tracing
Innovative multi-GPU acceleration for Vulkan-based ray tracing, enabling cloud offloading and dynamic workload balancing for enhanced rendering performance and scalability.