- VSAM structures
- VSAM commands
- Performance tuning
- JCL Buffers
- LSR Buffers
- System Buffers
- VSAM parameters
- IAM, a VSAM alternative
- VSAM Recovery
- VSAM RLS, DFSMStvs
System Managed Buffering (SMB) for VSAM datasets was introduced with DFSMS version 1.4 for KSDS files. This was enhanced with DFSMS 1.5 to include all VSAM files opened for NSR processing. Basically, the system decides how many buffers to use for data and index portions, and also whether to use NSR (best for sequential access) or LSR (best for OLTP acess). SMB will pick which buffering option is best for a specific job or application, and can even swap between them if necessary. SMB is better than the VSAM defaults, but not as good as third party buffering tools. This is because it is only invoked at file open time, whereas third party products will adjust the VSAM buffers as required over time.
The main restriction is that the data set must be in extended format, which means the data set must be SMS managed and use a data class that is defined with DSNTYPE=EXT.
Older releases of DFSMS could have a performance problem for DO processing as SMB allocated enough index buffers to hold the index at open time, but as the file grew, this buffer count was not enough. The only solution was to close and open the data set so that SMB recalculated the buffer count. Since DFSMS 1.11, SMB takes 20% for BUFNI, which is more than the previous implementation.
If you want to investigate how buffering is working, check out the SMF type 64 records that are generated when the file is closed. Also, look for an IEC161I message in the job or task output, as that will tell you which SMB buffering option was selected by VSAM for each opened data set.
The default for SMB is to allocate the buffer pool above the line, but you can change this if you wish.
With DFSMS V2.1, you can specify ACCBIAS and RMODE31 values in SMS data classes
You use Record Access Bias to tell VSAM how many and which type of buffers to use during batch processing, when accessing VSAM EF files. The Parameters available are:
If you specified direct optimization (DO), then you can further tune the way the buffers work with DD AMP parameters
There are two ways to invoke SMB
EXTENDED REC ACC
//ddname DD DSN=vsam.cluster.name,AMP=('ACCBIAS=SYSTEM'),DISP=SHR
Anything specified in the JCL will override the DataClass.
Other possible values for ACCBIAS are
These terms are all explained above
RMODE31 Specifies whether for VSAM to allocate the buffers and control blocks in 31-bit addressable storage. You can use this field independently of SMB. With SMB, the default location is in 31-bit addressable storage (above the 16-megabyte line). Without SMB, the default is in 24-bit addressable storage (below the line). The following values can be specified for RMODE31 in data class:
You may see a couple of other options quoted, CO (create optimised) and CR (create recovery). These are selected by the system, and cannot be specified
manually. They are used for initially loading a VSAM data set, CO is used with 'SPEED' and CR with 'RECOVERY'.
Three other 'AMP' JCL options are available to control how SMB uses buffers if ACCBIAS=DO is used. SMBVSP restricts the overall size of the buffers, SMBHWT will reserve hyperspace buffers, and SMBDFR can delay buffer writes until end-of-job, or until the buffer is full, whichever comes soonest. You specify the lot in an AMP statement as
These are just example numbers, you need to decide what is best for you. SMSVSP is nnK or nnM, SMBHWT is 0-99 and SMBDFR is Y or N
You can specify the SMB buffer processing as SO/SW/DO/DW as above, or simply use SYSTEM and let the system pick it out. It decides this based on the access method. SMB will use NSR buffers, unless DO is specified. It then changes the buffering internally to LSR. However, if SMB cannot get enough LSR buffers, it will change the buffering to DW.
The pecking order which decides if SMB will be invoked is JCL specifications, then the data class Record-Access-Bias parameter, then the MACRF values. That is, whatever is specified in the JCL will always take preference.