A curated selection of architecture insights and research releases we explored this week.
Infrastructure
🔹 Spanergy: Energy-Aware Distributed Tracing for Microservices
Introduces an energy-efficient approach to microservice monitoring that reduces the power overhead of distributed tracing, making it particularly relevant for cloud-native architectures.
Read the paper (EN)
🔹 ProFlow: RL-Based Proactive Flow Placement
Applies reinforcement learning to optimize network flow placement in data centers, improving overall throughput while reducing latency.
Read the paper (EN)
Architecture
🔹 Application-Driven Architecture Exploration for Cross-Layer Heterogeneous Systems
Explores application-centric architecture design techniques that enable more efficient utilization of heterogeneous computing resources across multiple system layers.
Read the paper (EN)
🔹 The Fabric Is the Cluster Driver: eBPF Policies for GPU-CXL Fabrics
Presents a novel resource management approach for GPU-CXL clusters using eBPF policies, improving both system performance and operational security.
Read the paper (EN)
Cloud Native
🔹 SmartGen: Seamless Disaggregated LLM Inference with Selective KV Cache Transfer
Introduces an innovative optimization technique for large language model inference that significantly improves throughput while reducing request latency.
Read the paper (EN)
🔹 QCOEM: Quantum Cloud Orchestration with Evolutionary Multi-Objective Optimization
Proposes an efficient orchestration framework for managing quantum computing resources in the cloud, expanding the possibilities for quantum applications.
Read the paper (EN)
Developer Tools
🔹 Gleam: An Adaptive CUDA API for GPU Sharing over LAN
Develops an API that enables efficient GPU resource sharing across networked devices, simplifying the development of distributed applications.
Read the paper (EN)
🔹 CAPS: Fine-Tuning CCA Timing
Introduces techniques for improving CCA timing characteristics, potentially delivering substantial performance gains for systems that rely on this mechanism.
Read the paper (EN)
AI & Machine Learning
🔹 PRISM: Evaluating POSIX Storage Systems for AI Research Workflows
Presents new methods for evaluating storage systems, helping optimize data-intensive AI research workflows.
Read the paper (EN)
🔹 Incast-Free MoE Rate-Based Scheduling
Introduces a scheduling strategy for Mixture-of-Experts systems that mitigates incast congestion and improves overall system performance.
Read the paper (EN)