Self-Driving Databases-It All Starts with Workload Forecasting

1 .

2 .▪ ▪ ➢ ▪ ➢ Improving Query Performance THE REAL WORLD OF THE DATABASE in Data Warehouses ADMINISTRATOR Business Intelligence Journal 06 Unisphere Research 15 

3 .▪ ➢ ➢ ▪ ➢ ➢ ➢ 

4 .Perception Action Model Planning 

5 .Perception Action Model Planning 

6 .

7 .▪ Prediction Horizon 1. Good Accuracy 2. Major Patterns 3. Cost vs. Accuracy Prediction Interval 

8 .▪ ▪ ▪ Cyclic (Diurnal) Growth and Spike 

9 . SQL SQL SQL Forecasts Query-based Workload Forecasting for Self-Driving Database Management Systems SIGMOD 18 

10 .▪ ▪ Millions → Thousands 

11 .▪ ➢ ➢ ➢ 

12 .

13 .A few large clusters exhibit major patterns 

14 .▪ ➢ ➢ ▪ LR+RNN has the best average accuracy 

15 .

16 .

17 .

18 .▪ ▪ ▪ ➢ ➢ 

19 .25% 

20 .▪ ▪ ➢ built ➢ building Benefit/Penalty Deployment Traditionally estimated by human 

21 .Perception Action Model Planning 

22 .▪ 

23 .▪ 

24 .▪ ▪ ➢ 

25 .Perception Action Model Planning 

26 .▪ ➢ ▪ ➢ ➢ 26 

27 .▪ 27 

28 .▪ ▪ ▪ ▪ ▪ 

29 .