Abstract: Database performance prediction has less historical data available and larger affection by external uncertainty factors. The traditional single, static prediction model is very difficult to meet the needs of actual production. This paper focuses on building a database dynamic prediction combination model under high load to overcome the deficiencies of the traditional Markov chain methods. So it can effectively predict the SQL performance problems caused by sharp increase of logical reads. The overall trends of random time series data can be predicted by building gray-Markov chain prediction model. When the Oracle database system is under high load conditions, these techniques can be used to accurately predict the executive performance of SQL statement, to quickly locate the performance bottlenecks and to effectively prevent the performance problems.

Key words: Oracle database, gray-Markov performance prediction, dynamic performance prediction model, combination model precision