Question

I read in a Microsoft T-SQL Performance Tuning whitepaper that correlated sub-queries can be costly in terms of performance on a large table:

…Compare this to the first
solution that would scan the whole table and execute a correlated subquery for every
row. The difference in performance is negligible on a small table. But on a large table it
may amount to hours of processing time…

Is there a general way to convert a query with several aggregations based upon different criteria as correlated sub-queries into a single query that uses JOINs instead of correlated sub-queries?

Consider an example:

Prepare the schema:

CREATE TABLE Student (
    ID INT NOT NULL PRIMARY KEY IDENTITY(1,1),
    Name NVARCHAR(255) NOT NULL
);

CREATE TABLE Grade (
    ID INT NOT NULL PRIMARY KEY IDENTITY(1,1),
    StudentID INT NOT NULL FOREIGN KEY REFERENCES Student(ID),
    Score INT NOT NULL,
    CONSTRAINT CK_Grade_Score CHECK (Score >= 0 AND Score <= 100)
);

INSERT INTO Student (Name) VALUES ('Steven');
INSERT INTO Student (Name) VALUES ('Timmy');
INSERT INTO Student (Name) VALUES ('Maria');
 
INSERT INTO Grade (StudentID, Score) VALUES (1, 90);
INSERT INTO Grade (StudentID, Score) VALUES (1, 81);
INSERT INTO Grade (StudentID, Score) VALUES (1, 82);
INSERT INTO Grade (StudentID, Score) VALUES (1, 82);

INSERT INTO Grade (StudentID, Score) VALUES (2, 99);
INSERT INTO Grade (StudentID, Score) VALUES (2, 63);
INSERT INTO Grade (StudentID, Score) VALUES (2, 97);
INSERT INTO Grade (StudentID, Score) VALUES (2, 90);

INSERT INTO Grade (StudentID, Score) VALUES (3, 66);
INSERT INTO Grade (StudentID, Score) VALUES (3, 61);
INSERT INTO Grade (StudentID, Score) VALUES (3, 60);

The query in question:

SELECT Name,
    (SELECT AVG(Score) FROM Grade WHERE StudentID = Student.ID AND Score < 65) AS 'F',
    (SELECT AVG(Score) FROM Grade WHERE StudentID = Student.ID AND Score >= 65 AND Score < 70) AS 'D',
    (SELECT AVG(Score) FROM Grade WHERE StudentID = Student.ID AND Score >= 70 AND Score < 80) AS 'C',
    (SELECT AVG(Score) FROM Grade WHERE StudentID = Student.ID AND Score >= 80 AND Score < 90) AS 'B',
    (SELECT AVG(Score) FROM Grade WHERE StudentID = Student.ID AND Score >= 90 AND Score <= 100) AS 'A'
FROM Student

Produces the following result:

Name    F     D     C     B     A
-----------------------------------------
Steven  NULL  NULL  NULL  81    90
Timmy   63    NULL  NULL  NULL  95
Maria   60    66    NULL  NULL  NULL

I am aware of the technique that you can use with COUNT() where you perform a single SELECT with a JOIN and then use a CASE statement to optionally add 1 to a counter when the primary keys line up between your join AND your condition is true. I am looking for a similar sort of technique that can be applied for different types of aggregations (as opposed to just COUNT).

Is there an effective way to convert this example query to use a JOIN instead of multiple sub-queries?

Answer 1

Maybe I’m missing something, but the solution using a CASE does work for aggregates as well:

SELECT st.name, 
       avg(CASE WHEN g.score < 65 THEN g.score ELSE NULL END) as F,
       avg(CASE WHEN g.score >= 65 AND g.score < 70 THEN g.score ELSE NULL END) as D,
       avg(CASE WHEN g.score >= 70 AND g.score < 80 THEN g.score ELSE NULL END) as C,
       avg(CASE WHEN g.score >= 80 AND g.score < 90 THEN g.score ELSE NULL END) as B,
       avg(CASE WHEN g.score >= 90 AND g.score <= 100 THEN g.score ELSE NULL END) as A
FROM Grade g
  JOIN Student st ON g.studentid = st.ID
GROUP BY st.name