In a cloud computing environment, a great amount of services require to be deployed into the cloud, which may lead to the redundancy and complexity of cloud-services, and ever-increasing resource consumption. Current researches try to optimize the increasing number of services with the support of buffer pool and deployment optimization algorithms. Buffer pool here performs the crucial task of storing the deployment requirements for optimization. However, issues like when to enforce the flush and which stored objects should be flushed out still lack comprehensive consideration. To address these problems, we propose an improved collaborative dynamic double buffer pool in this paper. By classifying different deployment requirements and flush fashions, we select a threshold to determine when to enforce the flush. According to this flush time, we can select the to-be-optimized objects dynamically along with necessary collaborative interaction with the service owner. Meanwhile we present the algorithms of the buffer pool flush mechanism, and the improved service deployment optimization approach. In the end, we illustrate the effectiveness and feasibility of our approach through a demonstration which guarantees relatively high optimization efficiency.