The paper presents the result of an evaluation of the performance of different message broker system configurations, which lead to the construction of the specific architecture guidelines for such systems. The examples are provided for an exemplary middleware messaging server software - RabbitMQ, set in high availability - enabling and redundant configurations. Rabbit MQ is a message queuing system realizing the middleware for distributed systems that implements the Advanced Message Queuing Protocol. The scalability and high availability design issues are discussed and the possible cluster topologies and their impact is presented. Since HA and performance scalability requirements are in conflict, scenarios for using clustered RabbitMQ nodes and mirrored queues are interesting and have to be considered with specific workloads and requirements in mind. The results of performance measurements for some topologies are also reported in this article.