Time Series Database Optimization Based on InfluxDB

The performance issues of InfluxDB’s TSM (Time-Structured Merge) storage engine, which is based on the LSMTree architecture, have been addressed in this paper. The focus is on eliminating write stalls, reducing write amplification, and optimizing resource utilization in scenarios with continuous longterm writes. An optimization design and implementation have been presented for InfluxDB to achieve these objectives. Moreover, to further enhance the modifications made to InfluxDB’s TSM storage engine, a compute-storage separated cluster architecture has been introduced and implemented. The objective is to tackle the native clustering solution’s lack of scalability. Experimental results demonstrate that the transformations applied to the InfluxDB storage engine in this paper have successfully achieved the design objectives, leading to improved write performance.