In multi -agent systems based on softbots, communication architectures have significant influences on system performance as interaction and cooperation of softbots are carried out via agent communication. In this paper we propose a new communication architecture which is based on a round-table mechanism. Communication channels are preliminarily defined based on the matching of agent requests. A channel connects an agent to a queue of matched agents with the same interests and is scheduled to become periodically active based on their proportions in total demand and the amount of available resources. The order of activating channels and the sequence of agents in matched queues are defined based on agent time constraints. Our evaluation shows that the proposed model achieves a good balance of performance and quality of service compared with the other methods and is especially useful when the number of agents are very large and the capacity of systems is limited.