• US & WORLD
• BUSINESS
• TECH
• ENTERTAINMENT
• HEALTH
• LIFESTYLE
• LOCAL

• Search for

• Random Article

 Home/Tech/Oracle’s MySQL HeatWave Starts Autopilot

Tech

Oracle’s MySQL HeatWave Starts Autopilot

danielkuhn1 day ago

The end of last year, Oracle Taken MySQL In a new direction by Introduction of cloud database (DBaaS) as a service It added analysis. Eight months later, the next action is to introduce a new “autopilot” feature. MySQL database service with HeatWave Use machine learning to improve database OLTP and OLAP performance. This is the second time Oracle has applied machine learning to optimize database operations, Autonomous databaseThe MySQL HeatWave autopilot feature is designed to bring intelligent automation to the market with small departmental databases that prefer open source, are cost sensitive, and are accustomed to manual knob turning.

Specifically, MySQL Autopilot uses internalized machine learning to assist some tasks of the user and applies closed-loop automation to handle other tasks. Assist users by making recommendations on how to provision and load databases, and implement closed-loop automation for failure handling / error recovery and query execution. Machine learning can also be used by Autopilot to iteratively improve query plans based on the results of other similar and different queries. In the latest release of MySQL HeatWave, the cluster size has been increased to 64, the capacity limit has been doubled to 32 TByte instances, and scalability has been increased by 20%.

For example, when provisioning a database, Autopilot uses adaptive sampling techniques to perform a smart scan of the table to predict the optimal cluster size. It is based on the amount of memory needed to hold the table and the prediction of intermediate results in memory. In this case, the output of the model makes recommendations, including a description of the recommendations to the customer who ultimately determine the sizing of the cluster.

Conversely, for query execution, the model uses logic to automatically schedule the query, reducing the conflict between short-term and long-run queries that is typical of mixed workloads. Without such smart query scheduling, short queries that are expected to be interactive can be delayed by long workloads unless the database administrator takes action. You can avoid this scenario by using Autopilot to predict the execution time of each query and prioritize short queries over long queries to reduce or eliminate waiting for interactive queries.

Other autopilot features include optimizations for loading and deciding where to physically place the data. Based on the query pattern, the algorithm attempts to place the most frequently joined rows closer within the same or adjacent partitions within the same node. Then, by comparing the proposed optimized placement with the defaults, you can predict the degree of improvement you can expect. Explaining the results helps customers decide whether to adopt HeatWave’s data placement recommendations.

A related feature that affects query planning is an improved automatic query plan that the system learns from queries and improves the performance of subsequent queries that are not necessarily identical to existing queries.

By comparison, query planners that use traditional relational databases are rule-based and tend to optimize cost or performance. Experienced DBAs then typically take on schema design tasks to optimize expected queries and lay out the data (loading data to optimize access). The MySQL Autopilot routine uses ML to automate these tasks. The company claims that the algorithm can improve the performance of benchmarks such as TPC-H (usually used for analysis) by up to 40%.