Research And Implementation Of Feedback-based Self Optimizing Concurrency Control

In the mobile internet era, enterprises need fast development to meet the market demandas to strengthen their market competitiveness. in this case, the database system design is easyto produce defects, leading to system instability or even catastrophic consequences whenfacing unpredictable high concurrency. Self optimizing concurrency control can deal with theunpredictable, however, the risk of completely self optimizing is too high to be implementedin the enterprise, commercial database self tuning tools are only to provide optimizationsuggestions for developers, it still need to research for a long time before realizing completelyself optimizing, at present, developers involved in tuning is the mainstream, how to reduce thethreshold to participate and implement concurrency control effectively is an important part ofself optimizing concurrency control.In view of this situation, this paper designs and implements a feedback-based selfoptimizing concurrency control model, which enables developers to participate tuning in aeasier way and ensure that the database system maintains stability in unpredictable highconcurrency. The idea of the model is to calculate transaction priority for concurrency controlby feedback info. Firstly. 5 system load indicators, respectively for the database connections,lock waiting, CPU, memory and IOPS is given to calculate the system load. Defines thetransaction conflict, transaction priority, priority effect, conflict description and adjustmentrules, introduces evaluation mechanism to avoid the interference of noise, and derives thetransaction priority formula. Then, implements the modules based on Redis kernel, introducesthe network communication protocol and how to minimize the cost of accessing to theexisting system.Finally, simulates two kinds of business, describes how developers can participate intuning with a simple configuration, using aliyun PTS to test the system in two ways: withoutfeedback mode and feedback mode. From the experimental results, the model can effectivelyguarantee the system’s stability in the scene under high concurrency, reduce importanttransaction failure rate, improve the overall system throughput.