Using DEFINE to Create SMF Data Sets

Before you can use an SMF data set, you must both define the data set and pre-format the data set. To define the data set, use the DEFINE access methods services utility. To preformat the data set with dummy records, use the IFASMFDP program. If you do not pre-format an SMF data set, SMF pre-formats it during initialization, which increases the time needed to IPL the system or to switch to a new, previously unused, SMF data set.

Defining an SMF data set: When you define an SMF data set, specify the following options for DEFINE:

REUSE indicates that the data set can be cleared by the dump program.
CONTROLINTERVALSIZE indicates the size of the SMF buffer. For the SMF data sets, you can specify ANY value between 0.5 (512 bytes) and 26K (26624 bytes), with certain restrictions. See Selecting the SMF Data Set Control Interval for more details.
SHAREOPTIONS has two sub-options (values) that define the level of sharing. The cross-region value must be 2. This indicates that sharing occurs with reading and serialization occurs with writing. The cross-system value is allowed to default.
NONINDEXED indicates that the entries are entry-sequenced.
SPANNED indicates that the records can span control intervals.
SPEED indicates that the data set will not be preformatted by VSAM while IFASMFDP is preformatting. (If SPEED is not selected, VSAM and SMF preformat concurrently.)

Note: SMF does not support extended-addressability VSAM data sets. Thus, the largest SMF data set cannot be larger than 4 gigabytes. For example, you must limit the number of cylinders you request to a maximum of 5800 for a data set on a 3390 device type.

If DATABASE 2 (DB2®) performance, serviceability, or audit data is sent to SMF, see Special considerations for DB2, JES3, and RMF.

Figure 1 shows the JCL statements needed to use the DEFINE utility to allocate one SMF data set on a direct access device and catalog it in the system catalog. This figure assumes that an IPL with the NOACTIVE SMF parameter was performed. Each SMF data set must be created according to this example before the first IPL that starts SMF recording.

Figure 1. Sample JCL Statements for Allocating the SMF Data Sets

//CREATE EXEC PGM=IDCAMS
//SYSPRINT DD SYSOUT=A
//SYSIN DD *
   DEFINE CLUSTER  (NAME(SYS1.MANX)             + 
                   VOLUME(xxxxxx)               + 
                   NONINDEXED                   + 
                   CYLINDERS(nn)                + 
                   REUSE                        + 
                   RECORDSIZE(4086,32767)       + 
                   SPANNED                      + 
                   SPEED                        + 
                   CONTROLINTERVALSIZE(nnnn)    + 
                   SHAREOPTIONS(2))
/*

Preformatting an SMF data set: When you use the IFASMFDP SMF dump program to preformat an SMF data set, use the BUFFERSPACE(81920) AMP=BUFND=nn option parameter on the DD statement specified in the INDD(ddname,OPTIONS(CLEAR)) parameter to improve the performance of the preformatting processing. (The number of data buffers desired is specified by nn, e.g., 10. Note that 10 buffers with a CISIZE of 26K will require 260K of memory.)

The step in Figure 2 preformats the data set. Using IFASMFDP to preformat the data set during the definition process avoids increasing the time required to IPL the system.

Figure 2. Sample JCL Statements for Preformatting the SMF Data Sets

//FORMAT EXEC PGM=IFASMFDP
//SYSPRINT DD SYSOUT=A
//NEWDS DD DSN=SYS1.MANX,DISP=SHR
//SYSIN DD *
   INDD(NEWDS,OPTIONS(CLEAR))
/*

Regardless of how the SMF data sets are created, the amount of DASD space they require depends on the amount of data generated and how often the data sets are dumped. The amount of data generated depends on the system work load and the record types selected for writing to the SMF data set.

If SMF data is available from a system similar to your own, you can use the report produced by the SMF dump program to estimate the amount of data generated. Otherwise, you might select a trial size for the data sets and adjust it as necessary. For example, you might start with two SMF data sets, each with 25 cylinders of space on a 3390. If the data sets fill up too quickly and data is lost, you can allocate more space for each data set, or create additional data sets.