Sunday, June 19, 2011


One of the more troubling aspects of DB2 database design and creation is the non-partitioning index (NPI). Creating NPIs on tables in a partitioned table space can pose management and performance issues. Partitioned table spaces tend to be large and by their very design will span multiple underlying data sets. Any partitioning indexes will also span multiple data sets. But what happens when you need to define non-partitioning indexes on a table in a partitioned table space?

The PIECESIZE clause of the CREATE INDEX statement can be used during index creation to break an NPI into several data sets (or "pieces"). More accurately, the PIECESIZE clause specifies the largest data set size for a non-partitioned index. PIECESIZE can be specified in kilobytes, megabytes, or gigabytes. For example, the following statement will limit the size of individual data sets for the XACT2 index to 256 megabytes:


Basically, PIECESIZE is used to enable NPIs to be created on very large partitioned table spaces. It breaks apart the NPI into separate pieces that can be somewhat managed individually. Without PIECESIZE, NPIs would be quite difficult to manage and administer. Keep in mind, though, that PIECESIZE does not magically partition an NPI based on the partitioning scheme of the table space. This is a misperception held by some. So, if you have a partitioned table space with 4 partitions and then create an NPI with 4 pieces, the data in the NPI pieces will not match up with the data in the 4 partitions.

When using PIECESIZE, more data sets will be created and therefore you can obtain greater control over data set placement. Placing the pieces on separate disk devices can help to reduce I/O contention for SQL operations that access NPIs during read or update processing. The elapsed time improvement may be even greater when multiple tasks are accessing the NPI.

Separating the NPI into pieces allows for better performance of INSERT, UPDATE and DELETE processes by eliminating bottlenecks that can be caused by using only one data set for the index. The use of pieces also improves concurrency and performance of heavy INSERT, UPDATE, and DELETE processing against any size partitioned table space with NPIs.

Keep in mind that PIECESIZE is only a specification of the maximum amount of data that a piece (that is, a data set) can hold and not the actual allocation of storage, so PIECESIZE has no effect on primary and secondary space allocation. Each data set will max out at the PIECESIZE value, so specifying PRIQTY greater than PIECESIZE will waste space. But also make sure that you avoid setting the PIECESIZE too small. A new data set will be allocated each time the PIECESIZE threshold is reached. DB2 will increment the A001 component of the data set name each time. Ideally, the value of your primary quantity and secondary quantities should be evenly divisible into PIECESIZE to avoid wasting space.

To choose a PIECESIZE value, divide the overall size of the entire NPI by the number of data sets that you wish to have. For example, for an NPI that is 8 megabytes, you can arrive at 4 data sets for the NPI by specifying PIECESIZE 2M. Of course, if your NPI grows over 8 megabytes in total you will get additional data sets. Keep in mind that 32 pieces is the limit if the underlying table space is not defined with DSSIZE 4G or greater. The limit is 254 pieces if the table space is defined as DSSIZE 4G or greater.

No comments: