Redis automatic performance tuning based on eBPF

In Internet applications, caching has become a key component of a high-concurrency architecture. As more and more companies use Redis as a caching database, the performance of Redis has become a particularly important part of the production environment. Performance tuning has always been a challenging field. Most engineers use the experience to determine the bottleneck of the system and then achieve performance tuning. This type of method is not only extremely time-consuming but also does not have the reusability of system parameters. In response to these problems, this article establishes an automatic tuning model that uses the eBPF program to identify different Redis working scenarios and uses the random forest method to sort the memory parameters in advance in different working scenarios to filter out the high-efficiency parameters. And then realize Redis automatic performance tuning by changing the parameter values. Different from the optimization at the application level, the operating system directly manages the resources of the hardware, so that not only the performance of the server can be maximized, but also the service scenario can be more suitable.