In edge collaborative computing, a single device can no longer support more and more complicated applications and services. Their tasks are offloaded to the adjacent edge server with rich computing and storage resources to match the various high calculation capabilities and low latency requirements. At the same time, the publish-subscribe system is commonly applied from the communication perspective to build a unified transmission protocol to protect data privacy. In the publish-subscribe system, tasks often execute periodicity, depend on each other and the data is distributed to different clients in high-frequency. However, the traditional task offloading algorithms mainly aim at the single data transmission and single task execution scenario, which cannot effectively handle the offloading characteristics in a publish-subscribe system. Therefore, we propose Mort, a task offloading and management framework. It supports task decomposition using the static program analysis technique, and the task dependencies are extracted and the parallelism is increased. Nonlinear integer programming-based modeling and group-based resource fusion-based offloading algorithms are applied to optimize network data transmission. The coordination process-based scheduling model is used to schedule the offloading tasks with less overhead. Our comprehensive experiments show that Mort’s data transmission optimization can reach 80%-90% of the optimal solution with a low overhead of only 2%.