Microsoft Showcases Networked Systems Innovations at NSDI 2024

Microsoft at NSDI 2024: Innovations in Networked Systems

This post highlights Microsoft's significant contributions and participation at the 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI '24), a premier event for researchers and experts in networked and distributed systems. Microsoft's involvement underscores its commitment to fostering innovation in this field, with team members playing key roles in organizing the event, including program committee and conference co-chair positions.

Key Contributions and Research Areas:

Microsoft researchers presented 19 papers covering a wide range of topics, demonstrating the breadth and depth of their work in networked systems. These papers span from early-stage research to systems already deployed in production, showcasing advancements in:

• Resource Management:
  • Autothrottle: A practical bi-level approach to resource management for microservices with Service Level Objectives (SLOs). This framework uses a learning-based controller to set performance targets for per-service heuristic controllers, achieving higher CPU savings and fewer SLO violations than existing solutions like Kubernetes.
• 5G and Wireless Networks:
  • Application-Level Service Assurance with 5G RAN Slicing: Introduces Zipper, a RAN slicing system that provides application-specific throughput and latency using a model-predictive control approach for optimal bandwidth allocation.
  • Spectrumize: Spectrum-efficient Satellite Networks for the Internet of Things: Proposes a novel technique using Doppler shift from satellite motion to improve packet detection and decoding for low-power IoT devices in Low Earth Orbit satellite networks, achieving significant performance gains.
• WAN and Cloud Infrastructure:
  • Solving Max-Min Fair Resource Allocations Quickly on Large Graphs: Presents algorithms that streamline max-min fair resource allocation for WAN traffic engineering and cluster scheduling, leading to quicker, fairer, and more efficient allocations, with a threefold increase in processing speed in Azure's WAN traffic engineering.
  • Making Kernel Bypass Practical for the Cloud with Junction: Focuses on making kernel bypass practical for cloud environments.
  • CHISEL: An optical slice of the wide-area network: Explores optical slicing for wide-area networks.
• Security and Anomaly Detection:
  • NetVigil: Robust and Low-Cost Anomaly Detection for East-West Data Center Security: An advanced anomaly detection system for data centers using graph neural networks and contrastive learning, outperforming existing solutions in accuracy, cost-efficiency, and speed.
  • Finding Adversarial Inputs for Heuristics using Multi-level Optimization: Addresses the challenge of understanding performance differences between heuristics and optimal solutions by identifying adversarial inputs that expose these gaps.
  • MESSI: Behavioral Testing of BGP Implementations: Focuses on testing Border Gateway Protocol (BGP) implementations.
• Machine Learning and AI Applications:
  • LitePred: Transferable and Scalable Latency Prediction for Hardware-Aware Neural Architecture Search: Enhances neural architecture search with transferable and scalable latency prediction.
  • Vulcan: Automatic Query Planning for Live ML Analytics: Automates query planning for live machine learning analytics.
  • Gemino: Practical and Robust Neural Compression for Video Conferencing: A neural compression technique for video conferencing that is practical and robust.
  • GRACE: Loss-Resilient Real-Time Video through Neural Codecs: Utilizes neural codecs for loss-resilient real-time video.
• IoT and Embedded Systems:
  • Cyclops: A Nanomaterial-based, Battery-Free Intraocular Pressure (IOP) Monitoring System inside Contact Lens: A novel system for monitoring intraocular pressure using a battery-free contact lens.
  • ADR-X: ANN-Assisted Wireless Link Rate Adaptation for Compute-Constrained Embedded Gaming Devices: Improves wireless link rate adaptation for gaming devices.
• System Monitoring and Optimization:
  • Sequence Abstractions for Flexible, Line-Rate Network Monitoring: Develops abstractions for flexible, line-rate network monitoring.
  • OPPerTune: Post-Deployment Configuration Tuning of Services Made Easy: Simplifies the process of tuning service configurations after deployment.

Key Awards and Recognition:

• Outstanding Paper Award: The paper on Autothrottle received this prestigious award, highlighting its significant impact and innovation in resource management.

Microsoft's Role in Organizing NSDI '24:

Several Microsoft researchers played crucial roles in the symposium's organization:

• Program Committee Co-Chair: Irene Zhang
• Program Committee: Paolo Costa, Anuj Kalia, Amar Phanishayee, Dan Ports, Francis Yan
• Mentoring Co-Chair: Jay Lorch
• Test of Time Awards Committee: Jay Lorch, Amar Phanishayee
• Steering Committee: Jay Lorch, Amar Phanishayee

Related Content:

The post also links to related publications, projects, and events, including:

• Publications: A comprehensive list of Microsoft's accepted papers at NSDI '24.
• Projects: MetaOpt, 6G | Space, Software-Driven Wide Area Networks, Project FarmVibes.
• Events: Microsoft at SOSP 2024, Microsoft at ASPLOS 2024, Abstracts: April 16, 2024, Microsoft at ACM SIGCOMM 2023.
• Labs: Microsoft Research Lab - Redmond.

Author and Social Media:

The post is authored by Ranveer Chandra, Managing Director, Research for Industry. It also includes links to share the post on various social media platforms (Facebook, X, LinkedIn, Reddit) and subscribe to RSS feeds.

Visuals:

The post features an image of the "Microsoft at NSDI 2024" logo and a link to "Azure AI Foundry Labs" with an associated image.

This summary covers the key highlights of Microsoft's participation and research presented at NSDI '24, emphasizing their advancements in networked systems, AI, and related fields.