Abstract:
We consider the task of storing sensor data that arrive in small
portions from a large number of sources. We suggest a general
sensor data storing model with RAM buffering and experimentally
evaluate its implementation based on Linux filesystems. We use
blktrace tool to study the slowdown that occurs with larger
number of sources, and discuss the ways to sustain performance. We
give experimental data for write performance and write amplification
in ext4 and f2fs filesystems.
(In Russian).
Key words and phrases:sensor data storage, Linux, filesystem, write performance, write amplification, time series data, TSDB.