Standard values for the DMS I/O rate

In section "Standard values for BS2000 servers" you will find guidelines for the recommended maximum I/O rate as a function of the CPU type. These values are governed by the rule that no more than a certain share of CPU capacity (for /390 servers 15 - 25% or for x86 servers 12 - 16% in processor state SIH+TPR) should be used for initiating and handling I/O operations. In individual cases, the recommended values may be considerably exceeded. This can by all means be a meaningful application. In such cases an individual check should be made to determine whether the violation of the recommended values causes performance problems.

ANALYZER reports for assessing the DMS I/O rate

Report group	Report	Meaning
IO	IOs for device classes	Number of DMS I/O operations/s
CATEGORY-IO	IOs for category	Number of DMS I/O operations/s per task category

Storage systems of type ETERNUS DX/AF can be monitored with the openSM2 Manager.

Tuning approaches for reducing the DMS I/O rate

The points below are not arranged according to general importance.

• Reorganize the ISAM files Lower the number of index levels by means of organizational measures (dividing the collected data, checking for data obsolescence).

• Switch to NK-ISAM files, with the resultant savings on I/O operations through the use of self-configuring ISAM pools or ISAM pools which can be selected on a user-specific basis. Use of the ISAM pool can be monitored with the aid of reports of the report group ISAM-POOL or ISAM-FILE:

  Report	Monitoring variable	Meaning
  Fix operations	Number of fix-hit operations	Number of ISAM accesses/s which were resolved from the ISAM pool
  	Number of fix-IO operations	number of ISAM accesses per second that resulted in an I/O operation to disk (read access)
  	Number of fix-wait operations	number of ISAM accesses per second during which the system had to wait for pool pages to become free
  Slot operations	Number of reserve slot wait operations	number of wait states per second resulting from bottlenecks in the internal pool management which were caused by the pool being too small
  Used pages	Number of total pages	number of PAM blocks in the ISAM pool which (after deducting the space required for management data) are available for buffering data
  Index operations	Number index operations	total number of index accesses
  	Number index hit operations	share of index accesses dealt with from the ISAM pool

  The ISAM pool is the correct size (for details of calculating the size, see the “DMS Macros” manual [8 (Related publications)]) when the following applies:

  #FIX-HIT OPERATIONS / (#FIX-HIT OPERATIONS + #FIX-IO OPERATIONS) >= 0,85

  The following must also be true:

  Number of fix-wait operations = 0 Number of reserve slot wait operations = 0
  
  In order to generate an ISAM pool that is of sufficient size, enough main memory capacity must be available to prevent the effect of reducing the number of DMS I/O operations being compensated for by an increase in the number of paging I/Os.

• Raise the blocking factor (for both disk and tape processing) Increasing the block size is always advisable in the event of sequential access. In the case of random-access processing (PAM, ISAM), this condition must be checked on an individual basis. Increasing the block size reduces the number of I/O operations and thus offloads the device and the CPU, but not the channel.

• Increase the size of the internal data buffer for database applications (please refer to the relevant database manuals) Whenever an increase in the size of the internal buffers leads to a corresponding increase in the size of the user address space, adequate main memory capacity must also be provided.

• Reduce the I/O rate to KDCFILE in UTM applications by increasing the size of the UTM cache area (see the “openUTM Administering Applications” manual [20 (Related publications)]).

• Use the software product DAB (ADM-PFA) By reading in up to 16 PAM blocks at a time (32 PAM blocks with AutoDAB), the number of I/O operations required is significantly reduced when processing is mainly sequential. Prerequisites are sufficient main memory and a corresponding CPU reserve capacity (CPU intensity increases when the physical I/O delay is reduced). See the “DAB” manual [5 (Related publications)] for information on using this function. The hit rate on the specified DAB subareas can be checked with the help of the reports “Reads for internal area” und “Writes for internal area” of the report group DAB. If AutoDAB is used, /SHOW-DAB-CACHING provides an overview of the files which are currently subject to caching. See also the section "ADM PFA caching".

• Use HIPERFILEs (see section "Working with HIPERFILEs and DAB").

If the common value for the DMS I/O rate is less than the suggested standard value (see the section "Standard values for BS2000 servers"), the next step is to check the workload on the channels and the attached peripheral devices.