RESIST: Resilient In-Network Computing

• Grant number: 23/00764-2
• Support Opportunities: Regular Research Grants
• Start date: February 01, 2024
• End date: January 31, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Agreement: MCTI/MC
• Principal Investigator: Alberto Egon Schaeffer Filho
• Grantee: Alberto Egon Schaeffer Filho
• Host Institution: Instituto de Informática. Universidade Federal do Rio Grande do Sul (UFRGS). Ministério da Educação (Brasil). Porto Alegre , SP, Brazil
• Associated researchers: Israat Haque
• Associated scholarship(s): 24/23855-6 - Development of a state replication and restoration system using high-level language, BP.TT; 24/05527-1 - Development of a mechanism to support different consistency models, BP.TT

Abstract

Recently, the possibility of using programmable forwarding devices (P4 switches) to allow not only the forwarding of packets but also the execution of computation within the network itself gave rise to a paradigm called in-network computing (INC). In-network computing allows offloading parts of a distributed system's functionality onto the programmable data plane. Examples of INC functionality that can be offloaded from traditional servers to the network include, for example, network caching, data aggregation, and concurrency control. The ability to perform computing tasks on the network brings several benefits, such as reducing the processing time of requests and the possibility of performing pre-processing and aggregation to reduce the amount of data transmitted over the network. However, data plane elements are subject to failures that can cause interruptions to the operation of INCs offloaded to the network and thus compromise the resilience of these applications. These failures can cause the loss of critical information and disrupt the operation of systems that rely on network functionality. This project's main objective is to investigate and design solutions to provide fault tolerance based on replication and consistent recovery of the behavior of INC applications running on programmable data planes, in order to make such applications more resilient. Among the additional contributions sought by this project, the following stand out: support for different notions of consistency to provide different correction semantics to the behavior of INCs; specification of a high-level language for specifying state restoration strategies; and models and heuristics for optimally allocating INCs and their replicas in the data plane. (AU)

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

NUNES, DIEGO CARDOSO; COELHO, BRUNO LOUREIRO; PARIZOTTO, RICARDO; SCHAEFFER-FILHO, ALBERTO EGON. No Worker Left (Too Far) Behind: Dynamic Hybrid Synchronization for In-Network ML Aggregation. INTERNATIONAL JOURNAL OF NETWORK MANAGEMENT, v. N/A, p. 18-pg., 2024-07-24. (23/00764-2, 20/05152-7)