| dc.description.abstract | In cloud system or big data analysis platform, Message Broker plays an important role in the message exchange of various system components. Generally speaking, the resources of each tenant should be isolated in multi-tenant cloud. However, Message Broker is still shared by many tenants on the same platform. When a certain tenants submit tasks with high computational complexity, the Message Broker resource will be consumed so heavily that the message delivery of other tenants cannot be processed immediately, resulting in lowering overall service quality of other tenants. The above-mentioned tenants who cause the Message Broker resources to be consumed heavily are called Bad Neighbor. In order to solve the situation that Bad Neighbor will occupy a large amount of Message Broker resources, we propose a dynamic bandwidth rate limiting allocation mechanism, which is suitable for multiple load balance. Under the load-balanced architecture, the Message Broker plugin and its API are used to find Bad Neighbor. Besides, Linux traffic control on the load balancing server is utilized to limit the speed of Bad Neighbor. For limiting rate of bandwidth allocation on each load balancing server, we use the network traffic flow on the current load balancing server since the user connection usage amount is not necessarily evenly distributed across the two load balancers. The network traffic flow is used to allocate the dynamic bandwidth speed limitation, which makes the Bad Neighbor able to complete the submission task more efficiently than the average speed-limiting bandwidth allocation method, even after being limited by rate. | en_US |