Latency-aware edge flow computation reinforcement learning partitioning method

Abstract

A latency-aware reinforcement learning-based partitioning method for edge stream computing belongs to the field of mobile edge computing technology. It aims to address the problem of optimizing processing latency in edge stream processing systems through partitioning. This invention includes establishing the state space of a data partitioning model and collecting the current state data of the data partitioning model; constructing a processing latency prediction model to generate rewards for the data partitioning decision model; establishing a data partitioning decision model, including designing the inputs and outputs, designing reward and training methods, and designing hotkey optimization methods; generating state data based on the state space of the data partitioning model to train the processing latency prediction model and the data partitioning decision model, resulting in a trained data partitioning decision model. This model scores each data partitioning decision based on the state data and selects the appropriate downstream operator instance based on the relative magnitude of the scores, thus completing one data partitioning decision.

Images