Message format pool tuning

The message format buffer pool holds message format blocks for frequently used transactions. Optimizing it reduces the I/O burden of common transactions.

To obtain storage in the message format buffer pool, the space holding the least recently referenced format block is freed so that blocks for transactions with high arrival rates are usually available. To minimize format block I/O, plan to have sufficient space in the pool to hold the input and output format block pairs for use by your most frequently used transactions

Using fetch request elements (FREs) significantly affects efficient use of the pool. Each FRE controls one format block in the pool. If all the format blocks are 1000 bytes long and 10 FREs are specified, the maximum pool space is 10000 bytes. FRE assignments are made in the BUFPOOLS macro during system definition, but can be overridden by the FRE parameter in the control region EXEC statement. You can use the /DISPLAY POOL MFP command to compare the amount of free space to the amount of total space. If the free space is consistently high, relative to the size of the pool, make sure that the FRE assignment is not preventing full use of the allocated storage.

Another way to improve the use of the message format buffer pool is enable application programs to share formats. For example, a generalized message input format can share common DIF/MID blocks with subsequent editing performed by Input Message Field Edit and Input Message Segment Edit exit routines.

A performance-related function is available with Message Format Service (MFS). You can use the MFS Service utility to generate a directory of direct pointers to the location of all or a selected subset of the format blocks. This index is resident in the message format buffer pool and minimizes the I/O required to look up the location of a format block in the active IMS.FORMATA/B library directory. (Each index entry requires 14 bytes.) Evaluate the trade-off between allocating index storage for at least the highly referenced blocks against the reduction of I/O per block.