Home   People   Publications  


Static Analysis on Programmable Data Plane: Towards Network Switch Parallelization [abstract]
Seungbin Song
Master's Thesis, Department of Creative IT Engineering, Pohang University of Science and Technology, June 2018.

Flexible and parallel packet processing is crucial for network switches to support various network services in high speed. While programmable data plane allows the flexible packet processing, lack of data dependence analysis on data plane programs limits parallelization of the packet processing. This thesis proposes static analysis and parallelization methods for the programmable data plane. The proposed analysis analyzes data dependence among packet processing tables, and finds pairs of data-independent tables. Furthermore, the proposed parallelization parallelizes the data-independent tables in full and staggered manners. This thesis evaluates the analysis and the parallelization on 14 sample P416 programs. The analysis finds 16 data-independent pairs among 37 tables, and the simulation of the parallelization reduces the number of stages in pipelines by 15.4%. This thesis will be useful to reduce latency of programmable network switches, and provide opportunities of parallelization and compiler optimizations.