Thursday, March 22, 2012

Working With Variable Data


One of the key design issues that is addressed in most DB2 applications is how to implement character data that varies in size from row to row. The basic trade-off is “ease-of-use and performance” versus “storage requirements.” It is possible to save storage space by using variable columns instead of placing small amounts of data in a large fixed space. Each variable column carries a 2-byte overhead, however, for storing the length of the data. Additionally, variable columns tend to increase CPU usage and can cause the update process to become inefficient. When a variable column is updated with a larger value, the row becomes larger; if not enough space is available to store the row, it must be moved to another page. This makes the update and any subsequent retrieval slower.

Follow these rules when defining variable character columns:
  • Avoid variable columns if a sufficient disk space is available to store the data using fixed columns. The general rule is: disk is cheap! Trading the cost of disk for more efficient development and subsequent performance is often an easy decision to make.
  • Though this may be obvious, do not define a variable column when the data does not vary from row to row.
  • Do not define a variable columns if its maximum length is less than 30 bytes. Furthermore, do not define a variable column if its maximum length is within 10 bytes of the average length of the column. Adhering to these two rules will cause you to choose VARCHAR data types only when they can potentially provide enough DASD savings to offset other costs.
  • Consider redefining variable columns by placing multiple rows of fixed length columns in another table or by shortening the columns and placing the overflow in another table.

If, after following these guidelines, VARCHAR columns need to be implemented, go ahead and do so. However, it is wise to continuously re-assess the decision to use variable character data.

So how can you monitor the effectiveness of variable columns? Well, it is possible to query the DB2 Catalog to determine the effectiveness of using VARCHAR for a column instead of CHAR. Consider, for example, the PROJNAME column of the DSN81010.PROJ table. It is defined as VARCHAR(24).

To gauge whether VARCHAR is appropriate for this particular column, consider issuing the following SQL query:

     SELECT    COL_LGTH, COUNT(*)
     FROM      (SELECT LENGTH(PROJNAME) AS COL_LENGTH
                FROM   DSN81010.PROJ
               )
     GROUP BY  COL_LGTH
     ORDER BY  COL_LGTH;


This query will produce a report listing the lengths (in this case, from 1 to 24, excluding those lengths which do not occur) and the number of times that each length occurs in the table.  These results can be analyzed to determine the range of lengths stored within the variable column. 

If you are not concerned about this level of detail, the following query can be used instead to summarize the space characteristics of the variable column in question:

    SELECT  24*COUNT(*),
            SUM(2+LENGTH(PROJNAME)),
            24*COUNT(*)-SUM(2+LENGTH(PROJNAME)),
            24,
            AVG(2+LENGTH(PROJNAME)),
            24-AVG(2+LENGTH(PROJNAME))
    FROM    DSN81010.PROJ;



This query produces a report similar to the one shown below:

SPACE      SPACE        TOTAL   AVG.       AVG.         AVG.
USED AS    USED AS      SPACE   SPACE AS   SPACE AS     SPACE
CHAR(24)   VARCHAR(24)  SAVED   CHAR(24)   VARCHAR(24)  SAVED
--------   -----------  ------  ---------  -----------  -----
158058        96515     61543     24          16          8



The following list itemizes the definition for each of the individual columns calculated by this query:

Definition                               
Calculation                                                 
Space Used As CHAR(24)
24*COUNT(*)
Space Used As VARCHAR(24)
SUM(2+LENGTH(PROJNAME))
Total Space Saved Using VARCHAR
24*COUNT(*)-SUM(2+LENGTH(PROJNAME))
Avg. Space Used As CHAR(24)
24
Avg. Space Used As VARCHAR(24)
AVG(2+LENGTH(PROJNAME))
Avg. Space Saved Using VARCHAR
24-AVG(2+LENGTH(PROJNAME))

The query can be modified to be used for any VARCHAR-defined column. The constant 24 can be changed to indicate the maximum length of the variable column as defined in the DDL. Using these tools, you can better judge the actual disk space savings accruing as a result of VARCHAR usage.

Synopsis

There are sound reasons for using variable data types within DB2 databases. Yet, it is important to remember that business conditions change and what may have been a sound reason for using VARCHAR data in the past, may no longer be sound. Use the queries in this article to determine if it is still appropriate to use VARCHAR data. In general, do not use VARCHAR for small columns or for columns whose length does not vary considerably.

Sunday, March 04, 2012

Fetching Multiple Rows

When you need to retrieve multiple rows, consider deploying a multi-row fetch to transfer more than one row using a single FETCH statement. This capability was added as of DB2 Version 8.

A multi-row FETCH retrieves multiple rows at one time into column arrays in your application program. By fetching multiple rows at once, your request can become more efficient, especially for distributed requests. The performance improvement using multi-row FETCH in general depends on several factors, such as whether the request is distributed, the number of rows to be fetched, the complexity of the SELECT statement, and the number of columns being fetched.

Nevertheless, using multi-row FETCH (in local environments) can improve performance with a significant reduction of CPU time possible. Tests conducted by IBM have shown performance gains of between 25% and 40% processing 10 rows per SQL statement for programs processing a considerable number of rows. With such significant gains possible, why hasn’t everyone moved to multi-row FETCH? Well, perhaps because it requires programming changes. A multi-row FETCH requires a cursor defined with rowset positioning. A rowset is a group of rows that are operated on as a set. Such a cursor enables your program to retrieve more than one row using a single FETCH statement. By fetching multiple rows at once, your request might become more efficient, especially for distributed requests.

 To use this feature, you must DECLARE your cursor as using the WITH ROWSET POSITIONING parameter. For example

EXEC SQL
  DECLARE CURSOR SAMPCURS
  WITH ROWSET POSITIONING
  FOR
  SELECT DEPTNO
  FROM   DSN81010.DEPT
END-EXEC.

Furthermore, to use a multi-row fetch you must have defined the appropriate structures to receive multi-row data. This means you must have defined an array of host variables into which the fetched rows can be placed. Each column fetched requires its own host variable array into which its values will be placed. Be sure to match the array size to the rowset size. With the appropriate setup coded, FETCH statements can be written to retrieve more than a single row from the result set. For example

FETCH ROWSET FROM SAMPCURS
  FOR 5 ROWS
  INTO HOSTVAR-ARRAY;

As you can see, the multiple-row fetch block is identical to a single-row-fetch block, except that there are two additional clauses—ROWSET and FOR n ROWS. The ROWSET clause specifies that the orientation of this cursor is rowset positioning (instead of single row). The FOR n ROWS clause specifies the size of the rowset to be returned. The maximum rowset size is 32,767.

Rowset cursors are very useful when you need to retrieve many rows or large amounts of data in distributed systems. By retrieving multiple rows with a single FETCH, multiple trips between the application and the database can be eliminated, thereby improving network performance.

To learn more about multi-row FETCH consider attending my upcoming webinar on the topic. This presentation will introduce and define multi-row FETCH, how to use it, and the performance implications of doing so. The presentation will also touch upon multi-row UPDATE. And it will introduce the new SoftBase Attach Facility MRF Feature, which allows you to implement multi-row FETCH without coding changes. To attend, sign up at this link: https://www1.gotomeeting.com/register/754473201