A Comprehensive Survey of Machine Learning in Scheduling of Transactions

As the number of core processors continues to increase, the current main-memory database system architecture becomes more challenging. The performance of the system can be improved if transactions are scheduled effectively, such that it can help avoid conflicts. A supervised or unsupervised machine learning model can be used to schedule the transactions that can estimate the probability of conflict. In this paper, the design of several intelligent transaction scheduling algorithms that are effective in improving the performance of the system are reviewed. The need for machine learning (ML) in database management systems (DBMS) and Transaction Processing Systems implementing online transaction processing (OLTP) has been strongly motivated in this paper. To inspire the need for this, the major focus of this paper is on employing artificial intelligence (AI) in systems to enhance their performance and also incorporate adaptive learning methods like intelligent scheduling or smart concurrency control methods to reduce response-time, handle or remove aborts and increase the throughput.