In order to meet the performance requirements and reduce resource consumption, researchers have proposed many autoscaling schemes. However, most of them only considered current states of the servers or application that limits the applicability of autoscaler. This paper presents an application-oriented autoscaler based on Long Short Term Memory network and Multi-Layer perceptron. The autoscaler includes a workload prediction model, a response time prediction model and a resource adjusting model. With these models, workload and response time of service are accurately predicted, and mean absolute percentage error of workload prediction is reduced to 3.3 × 10 -4 . The autoscaler can also provide appropriate adjustment suggestion to server managers. The experiments compared our autoscaler with other models, and results show that our autoscaler had better performance in workload prediction and server adjusting.