SmartNICs in the Cloud: The Why, What and How of In-network Processing for Data-Intensive Applications
Open Access
- 9 June 2024
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 556-560
Abstract
In modern datacenters and clouds, Resource Disaggregation has been adopted as a way of offering scalability and efficient resource utilization for large-scale applications. Provisioning CPU, memory, and storage resources independently for distributed data-intensive applications is a great enabler but it also brings challenges, especially in the form of networking and processing overhead. To reduce this overhead and to make disaggregation-related tasks significantly more efficient, cloud providers are offloading these to the network, i.e., to Smart Network Interface Cards (SmartNICs) and Smart Switches. Beyond this specific use-case, the presence of such programmable hardware in commercial clouds creates future opportunities for offloading application-level operations, e.g., parts of SQL queries or ML pipelines. To map out this exciting space, in this tutorial we take a detailed look at SmartNICs, explaining how they work, giving examples of what they are good for, and highlighting how they can best be utilized to make future data-intensive and distributed systems more efficient.Keywords
This publication has 16 references indexed in Scilit:
- A Cloud-Scale Characterization of Remote Procedure Calls2023
- The AMD 400-G Adaptive SmartNIC System on Chip: A Technology PreviewIEEE Micro, 2023
- The evolution of Amazon redshiftProceedings of the VLDB Endowment, 2021
- A new golden age for computer architectureCommunications of the ACM, 2019
- The Datacenter as a Computer2019
- CaribouProceedings of the VLDB Endowment, 2017
- Low-latency TCP/IP stack for data center applications2016
- Flash storage disaggregation2016
- Profiling a warehouse-scale computer2015
- QuasarACM SIGPLAN Notices, 2014