| CPC G06F 16/217 (2019.01) [G06N 3/04 (2013.01); G06N 3/08 (2013.01); G06N 20/00 (2019.01)] | 13 Claims |
|
1. A database performance tuning method, performed by a computer device, the method comprising:
receiving a performance tuning request of tuning a configuration parameter of a target database, the target database being a cloud database of a cloud service provider;
invoking a standard deep reinforcement learning model, the standard deep reinforcement learning model being trained with standard database instances and comprising a first deep reinforcement learning network and a second deep reinforcement learning network, the first deep reinforcement learning network being configured to provide a recommendation policy for outputting a recommended configuration parameter according to a status indicator, the second deep reinforcement learning network being configured to evaluate the recommendation policy provided by the first deep reinforcement learning network;
replaying an actual workload of the target database by: returning, by a load generator, the target database to a state at a previous timestamp; and re-executing a plurality of operations logged in an operation execution record of the target database within a historical time starting from the previous timestamp according to a same execution sequence logged in the operation execution record;
performing at least one round of retraining process on the standard deep reinforcement learning model in a process of running the target database according to the actual workload;
stopping retraining the standard deep reinforcement learning model when a training stop condition is met, to obtain the tuned deep reinforcement learning model;
obtaining a status indicator of the target database; and
inputting the status indicator of the target database into the tuned deep reinforcement learning model, to obtain a recommended configuration parameter of the target database,
wherein the target database tuned based on the recommended configuration parameter has a higher concurrency and a lower latency compared to the target database before tuning, the concurrency indicating a quantity of requests processed by the target database per unit time.
|