How Oracle DBAs Tackle Performance: Real‑World Stats Tuning and Career Insights

Host "PiPi" welcomes Shi Yuedong, Deputy General Manager and Database Technology Director at Beijing Blue Dawn Information Technology, to the 93rd episode of the interview series. Shi introduces himself as a 2006 computer‑science graduate who began working on data‑warehouse projects, later moving to Beijing and accumulating eight years of Oracle experience.

Career Milestones and Certifications

Shi earned PMP, Oracle 10g and 11g OCM certifications, co‑founded Blue Dawn in 2017, and led Oracle training and outsourcing services. He emphasizes that his technical path emerged from seizing opportunities and continuous learning.

Philosophy on Opportunity and Preparation

Drawing on Engels' quote, Shi argues that opportunities favor the prepared mind; diligent skill‑building and patience are essential for success.

Personal Interests: Writing a Novel

During preparation for the 10g OCM exam, Shi wrote a twelve‑day novel titled "Victim's Confession" and published it on Tencent's Yunqi Academy, later declining a contract to remain a free writer.

Performance Diagnosis Case at Lenovo

Shi describes a real‑world issue where a daily stored procedure at Lenovo Shenzhen suddenly slowed to an hour. Initial checks of locks, execution plans, and indexes revealed no problems. The root cause was outdated table statistics: the collected record count differed from the actual count.

He discovered the statistics were gathered using DBMS_STATS.GATHER_TABLE_STATS with estimate_percent => DBMS_STATS.AUTO_SAMPLE_SIZE. Changing estimate_percent to 100 (full sampling) resolved the slowdown.

Experiment to Determine AUTO_SAMPLE_SIZE Value

Using Oracle 12.1, Shi performed a series of commands to measure execution time for different estimate_percent values, illustrating that NULL (full sampling) and 100 yield similar times, while the default AUTO_SAMPLE_SIZE behaves like a value around 65%.

SQL> create table t_val (name varchar2(30), value number);
SQL> begin
  insert into t_val values ('AUTO_SAMPLE_SIZE', DBMS_STATS.AUTO_SAMPLE_SIZE);
end;
/
SQL> select * from t_val;
NAME                        VALUE
--------------------------- -----
AUTO_SAMPLE_SIZE               0

SQL> exec dbms_stats.gather_table_stats(ownname => 'SYS', tabname => 'T', estimate_percent => DBMS_STATS.AUTO_SAMPLE_SIZE);
-- execution time: 0.608 seconds

SQL> exec dbms_stats.gather_table_stats(ownname => 'SYS', tabname => 'T', estimate_percent => 100);
-- execution time: 0.889 seconds

SQL> exec dbms_stats.gather_table_stats(ownname => 'SYS', tabname => 'T', estimate_percent => 65);
-- execution time: 0.608 seconds

The results indicate that in Oracle 12.1 the default sampling percentage is approximately 65%.

Views on Big Data and Data Warehousing

Shi reflects on why data warehouses never became mainstream: fragmented enterprise systems, insufficient data volume and quality, immature internet business models, and limited data‑mining techniques at the time. He cautions that big‑data hype may lead to a near‑future plateau, recommending a calm, studied approach.

Conclusion

Shi emphasizes that success is not defined by wealth or fame but by personal integrity and steady effort. He encourages IT professionals to keep learning, stay prepared, and seize opportunities when they arise.