User:Magical Squirrel 001/sandbox
Programmable Data Plane Intelligence
[edit]Programmable data plane intelligence refers to achieving networking for machine learning (ML) and ML for networking with programmable data plane[1]. Programmable data plane intelligence summarizes the mutual integration and enabling relationship between programmable data planes and machine learning.
Programmable data plane (PDP) means that the network device's low-level packet processing logic is systematically, rapidly, and comprehensively reconfigured. In terms of network devices, programmable data planes include programmable switches and SmartNICs.
History
[edit]The entire network becomes a fully programmable platform after network interface cards, switches, and protocol stacks are programmable. Wai-xi Liu, Yong Cui, et al. first proposed programmable data plane intelligence in 2022[1].
In recent years, a series of in-network X works have emerged in programmable data plane intelligence.
- In terms of machine learning for networking, compared with deploying ML on control planes or end-hosts, deploying ML on PDP shows some unique advantages, including shorter decision delay and less decision overhead—for example, in-network defense, in-network classification, in-network clustering, and in-network load balancing.
- In terms of networking for machine learning, compared with other network technologies, PDP can reduce training overhead and time for ML—for example, in-network aggregation and in-network neural networks.
- ^ a b Liu, Wai-xi; Liang, Cong; Cui, Yong; Cai, Jun; Luo, Jun-ming (2022). "Programmable Data Plane Intelligence: Advances, Opportunities, and Challenges". IEEE Network: 1–8. doi:10.1109/MNET.124.2200113. ISSN 0890-8044.