Configuration Automatic Tuning for Distributed Message Systems by Exploiting Historical Knowledge

Currently, distributed message systems (DMSs) are widely used in large-scale systems to assist system functionality and continuously bring performance benefits. Usually, the performance of a DMS system is highly dependent on its parameter configuration. However, due to its hundreds of parameters far beyond manual labor, default configurations are usually adopted, resulting in poor performance. The previous DMS configuration tuning proposals suffer from long tuning times due to the huge search space. To address this issue, this paper proposes an efficient DMS configuration automatic tuning scheme named DCT-HK, that can achieve tuning goals within a limited time budget. The key design is to leverage the historical tuning knowledge to reduce search space, thereby the time cost to find the optimal configuration can be significantly reduced. The experimental results validate the effectiveness of DCT-HK. Under the same tuning time budget, our scheme achieves better tuning performance than baseline methods.