Using Statistical Analysis to Determine Causes of DBMS Performance Degradation – Part 2
Important detail: the distribution of performance and load on the DBMS is symmetrical. The load is constant.
14:41 – 15:11 – Performance Changes
Important detail: The load is constant.
15:11 – 15:45 – Low productivity
Important detail: the distribution of performance and load on the DBMS is symmetrical. The load is constant.
Conclusions
The load on the DBMS during the period of low DBMS performance (15:11-15:45) – has not changed
During the period of change in DBMS performance (14L41-15:11) – reverse correlation between the number of active sessions and DBMS performance – absent. Or in other words – a decrease in the load on the DBMS correlates with a decrease in the performance of the DBMS.
The distribution of DBMS performance during periods of high load (14:05-14:41) and low load (15:11-15:45) is symmetrical.
2. Analysis of DBMS expectations and performance
14:05 – 14:41 – High performance
14:41 – 15:11 – Performance Changes
15:11 – 15:45 – Low productivity
Conclusions
The correlation and composition of expectations during the period of high (14:05-14:41) and low (15:11-15:41) productivity do not completely coincide
The total weight of wait events during the low performance period (15:11-15:41) is significantly lower than during the high performance period (14:05-14:41).
Conclusion and general results
The decrease in DBMS performance in the period 15:11 – 15:45 was caused by a change in the nature of requests to the DBMS and is not an incident of DBMS performance degradation.
The use of statistical analysis has made it possible to dramatically reduce the time required to find possible causes of a performance incident.
A symmetrical distribution of DBMS performance can be considered with some degree of confidence as a sign of normal DBMS operation.
