Abstract: To improve the distributed system performance which is affected by load unbalance， this paper proposes and implements an impulsive and switching load balancing model with time delay based on control theory. The model constructs the corresponding sub-system according to the dynamics of node resources. The overloading is migrated by the load migrate rule， of which the migrate proportion is calculated by real-time running states of nodes. Only on this moment， the node broadcasts its message to others. Thus， the communication cost among nodes decreases， and the efficiency of dynamic load balancing increases. This paper provides the corresponding load balancing algorithm and conducts the evaluation on a real platform. Experimental results demonstrate that compared with other load balancing algorithms， the load balancing time of the proposed model is reduced by 29.82% on average.

Key words: load migration, node state, impulsive and switching system, dynamic load balancing, distributed system