TechTip: DB2 for i Optimization Strategies, Part 3

DB2
Typography
  • Smaller Small Medium Big Bigger
  • Default Helvetica Segoe Georgia Times

Learn ways to keep your database code lean and mean, resulting in better performance.

 

This third tip in the series continues where Part 1 and Part 2 left off. Presented herein are some general guidelines for how to make your database code purr like a kitten.

 

Combine Individual Insert Statements

I often see developers do things in SQL one step at a time, when the steps can be combined into a single statement. Consider the following code that inserts data multiple times into the same table:

 

INSERT INTO SPECIAL_CHARGES

(SalesOrderId,ChargeDesc,ChargeAmount)

VALUES(@SalesOrderId,'Taxes',@TaxAmount);

INSERT INTO SPECIAL_CHARGES

(SalesOrderId,ChargeDesc,ChargeAmount)

VALUES(@SalesOrderId,'Freight',@FrtAmount);

INSERT INTO SPECIAL_CHARGES

(SalesOrderId,ChargeDesc,ChargeAmount)

VALUES(@SalesOrderId,'Handling',@HndAmount);

 

These three INSERTs can be combined into one:

 

INSERT INTO SPECIAL_CHARGES

(SalesOrderId,ChargeDesc,ChargeAmount)

VALUES

(@SalesOrderId,'Taxes',@TaxAmount),

(@SalesOrderId,'Freight',@FrtAmount),

(@SalesOrderId,'Handling',@HndAmount);

 

According to my SQL monitor, DB2 processed the tri-row INSERT statement in about the same amount of time as one of the single-row statements. Prior to IBM i 6.1 and the VALUES clause, a UNION ALL operator could be used to create a single result set as well:

 

INSERT INTO SPECIAL_CHARGES

(SalesOrderId,ChargeDesc,ChargeAmount)

SELECT @SalesOrderId,'Taxes',@TaxAmount

FROM SYSIBM.SYSDUMMY1

UNION ALL

SELECT @SalesOrderId,'Freight',@FrtAmount

FROM SYSIBM.SYSDUMMY1

UNION ALL

SELECT @SalesOrderId,'Handling',@HndAmount

FROM SYSIBM.SYSDUMMY1

 

In this case, the UNION ALL statement is only slightly faster than the three individual insert statements. 

 

With simple examples like this, it’s not likely to make a big impact on the system, and users will not notice a small fraction of a second difference. However, as I mentioned in Part 1 about combining multiple update statements into one, think about the many things DB2 has to deal with when changing table data:

  • Security authorization check
  • Constraint processing (primary key, foreign key, unique index, check constraint)
  • Update of secondary indexes
  • Trigger processing
  • Checking for locks held by other processes
  • RCAC processing (optional)
  • Transaction processing (optional)
  • Generating journal entries (optional)

 

Why make DB2 for i do the work several times instead of once? Coding correctly, even in trivial cases, will foster good habits as developers realize the implications of their code in the eyes of DB2.

 

One more thing: Not every group of INSERTs into the same table should necessarily be combined into one statement for optimization purposes. If the data sets being inserted are large, there could be a decrease in performance for other users due to DB2 holding locks for a longer period of time (potentially blocking other processes, etc.) due to DB2 having to take longer to sort the combined rows before inserting into the table, etc.

 

Avoid Operations That Cause Table or Index Scans

In contrast to the prior tip about combining several steps into one, this tip involves breaking apart a complicated SQL statement into simpler parts that are easier for DB2 to digest.

 

Consider this seemingly innocuous query that gets the order detail for a parent order and companion child orders (with SalesOrderId as the high order primary key column):

 

SELECT *

  FROM SalesOrderDetail

WHERE SalesOrderId=@ParentOrderId

   OR SalesOrderId IN (SELECT ChildOrderId FROM Session.ChildOrders)

 

The WHERE predicates contain two simple requests that should be easy for DB2: Get the detail for the parent order (stored in a variable) and any child orders (stored in a temporary table). However, because of the OR operator, DB2 has to scan through all the order IDs to see if there is a match with the requested parent order ID or child order IDs.

 

There are 121K rows in the SalesOrderDetail table, and DB2 decided to scan one of the table’s smaller indexes to find which order numbers are a match, as shown in the Visual Explain plan in Figure 1a:

 

052716SansoterraFigure1a

Figure 1a: In this Visual Explain representation of the steps taken to access the data, notice how the OR in the WHERE clause forces DB2 to scan all of the order IDs in one of the table’s secondary indexes (top right index scan operator).

 

In contrast, if the query is broken into two segments (one for each search type: parent order ID and child order IDs), the resulting query actually reduces the work for DB2:

 

SELECT *

  FROM SalesOrderDetail

 WHERE SalesOrderId=@ParentOrderId>

UNION ALL

SELECT *

  FROM SalesOrderDetail

 WHERE SalesOrderId IN (SELECT ChildOrderId FROM Session.ChildOrders)

 

The reason that this type of query is better for DB2 is because the simplified WHERE predicates (without an OR) allow DB2 to access the data by the table’s primary key for just the requested SalesOrderIds without having to read through every SalesOrderId. The improved access plan is evidenced by the simplified Visual Explain diagram:

 

052016SansoterraFigure2b

Figure 1b: This query plan is much simpler than its predecessor because, even though there are two separate queries combined with a UNION ALL, DB2 can use the table’s primary key to get the information it needs without having to scan through all of the sales order IDs.

 

Between these two scenarios, on average, the second version of the query performed 2–3 times faster than the first one. The speed improvement can be even more dramatic when the queries are run the first time or when there is a greater volume of data.

 

Alternatives for simplifying the query would be to add the parent ID to the temporary table of child IDs (if possible) or to combine the child order IDs with the parent ID in a single subquery as follows:

 

SELECT *

  FROM SalesOrderDetail>

 WHERE SalesOrderId IN

(SELECT ChildOrderId FROM Session.ChildOrders

 UNION ALL

 SELECT * FROM (VALUES(@ParentOrderId)) D(ParentOrderId)

);

 

Refactoring Reaps Benefits

When refactoring database code, as a general rule, I will combine smaller steps (in terms of work done by a single SQL statement) into a single step where possible (as shown in the first example). But for medium and larger statements, I’m more cautious about combining them in the manner shown above (and if I don’t have time to test my changes compared to the original, I leave existing code alone!). In contrast, sometimes SQL statements doing a large amount of work can be broken into multiple statements or rewritten to allow DB2 to process the data more efficiently.

 

Optimize, Optimize, Optimize

These code samples are presented to demonstrate how certain coding practices can make DB2 work harder than it needs to. Whether preventing a trigger from needlessly firing multiple times or preventing DB2 from doing a full index scan (often a waste of I/O and memory), good coding practices lead to better-performing applications.

 

 

BLOG COMMENTS POWERED BY DISQUS