In-Network Offloading Approaches for Network Functions and Applications: Evaluation and Optimization

• Grant number: 24/21866-0
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Start date: March 24, 2025
• End date: March 23, 2026
• Field of knowledge: Engineering - Electrical Engineering - Telecommunications
• Principal Investigator: Christian Rodolfo Esteve Rothenberg
• Grantee: Francisco Germano Vogt
• Supervisor: Chrysa Papagianni
• Host Institution: Faculdade de Engenharia Elétrica e de Computação (FEEC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Institution abroad: University of Amsterdam (UvA), Netherlands
• Associated to the scholarship: 24/05516-0 - In-network computing optimization techniques, BP.DR

Abstract

In-network computing (INC) has become a key paradigm for enhancing the performance of modern data centers by offloading computational tasks to programmable devices like SmartNICs and switches. This approach offers benefits such as higher throughput, reduced latency, and improved power efficiency, which are crucial for applications that require strict performance metrics. Whileexisting network-level telemetry systems, such as in-band network telemetry (INT), provide valuableinformation, recent advances have shown that combining network-level metrics with application- and server-level data leads to more efficient resource orchestration and management. Our proposal focuses on two key aspects of INC. First, we aim to improve the monitoring of networks, servers, and applications by enhancing the exchange of information between servers/applications and network devices. This integration not only boosts the performance of existingin-network applications but also enables the development of new ones. Second, we investigate potential use cases for offloading, including the orchestration of intra-server resources such as Virtual Network Functions (VNFs). By leveraging SmartNICs, this approach facilitates real-time traffic optimization and resource allocation, reducing CPU usage and improving power efficiency. Addressing challenges like the limited computational power of SmartNIC's processors and synchronizing VNF metrics between hosts and SmartNICs will be central to ensuring scalability, resilience, and improved performance in modern data centers.