A selection of architectural insights and releases we’ve been reading this week.
Infrastructure
🔹 Robust Constraint-Aware Bayesian Tuning of BBRv2 for QUIC under Tactile Internet Constraints
A new approach to tuning BBRv2 for QUIC is proposed, taking into account the constraints of the Tactile Internet. This can improve performance under conditions of high latency and unstable connectivity.
Read the release (EN)
🔹 Scaling 5G-TSN Bridges
This study explores operating modes, scheduling, and time synchronization for 5G-TSN bridges under heterogeneous industrial traffic, which is critical for ensuring reliability and performance in modern networks.
Read the release (EN)
🔹 Vorticity Dissipation Based Routing
A new routing methodology based on vorticity dissipation is proposed, enabling infinite transport in ultra-dense networks and improving data transmission efficiency.
Read the release (EN)
Cloud Native
🔹 From Reactive to Autonomous: Evolution of AI Operations in Cloud Network Infrastructure
The study discusses the transition from reactive to autonomous AI operations management in cloud network infrastructure, helping to improve efficiency and reduce maintenance costs.
Read the release (EN)
🔹 SLO-Scaler: Uncertainty-Aware SLO-Driven Autoscaling for Microservices
A new autoscaling mechanism for microservices is introduced that accounts for uncertainty in SLO compliance, enabling more efficient resource management.
Read the release (EN)
🔹 The Lazy Pod That Lies
The study examines lazy-loading semantics for container images used to serve models on Kubernetes, helping optimize resource utilization and reduce idle time.
Read the release (EN)
Databases
🔹 Generalizing and accelerating consistency checking for non-transactional distributed storage systems
A new approach to consistency checking in non-transactional distributed storage systems is proposed, significantly accelerating the process and improving data reliability.
Read the release (EN)
🔹 Discovering KV Cache Eviction Policies via LLM-Guided Program Evolution
Using LLMs for program evolution makes it possible to discover optimized cache eviction policies for key-value storage systems, improving performance and memory efficiency.
Read the release (EN)
Developer Tools
🔹 AWS Open-Sources Dogwood
AWS has open-sourced Dogwood, an extension of Cedar designed to manage sequences of agent tool calls, simplifying the development and management of complex automation scenarios.
Read the release (EN)
🔹 Adaptive Snapshots Require Visible Reads
The study explores the need for visible reads in adaptive snapshots, which could significantly improve the performance of systems that use snapshots for state management.
Read the release (EN)
AI & Machine Learning
🔹 Could Model Partitioning Make Federated Learning More Sustainable?
The study explores whether model partitioning can make federated learning more sustainable by reducing computational costs and improving the efficiency of distributed systems.
Read the release (EN)
🔹 Empirical Evaluation of Cross-Carrier MCPTT & OTT MCX Interoperability in High-Density Environments
An empirical evaluation of MCPTT and OTT MCX interoperability in high-density environments is presented, which is critical for ensuring reliable communications in emergency situations.
Read the release (EN)
This week brought a number of significant research findings and developments that could shape the future of network infrastructure, cloud technologies, and software development.