What are the performance considerations for Data Compression doing CA Disk Backup or Archive to disk
1) Specify Data Compression via DCDATACPy only for ARCHIVE/BACKUP to DASD
Make sure that Data Compression via DCDATACPy is not used for ARCHIVE/BACKUP to TAPE. For example, set DCDATACPN in the SYSPARMS member and override this in all ARCHIVE/BACKUP to DASD job steps with DCDATACPY via //SYSPARMS DD * override in the job.
2) Consider using DCCOMPTC2 (compression technique 2)
Consider to use DCCOMPTC2 (compression technique 2) instead of the default setting DCCOMPTC0 (compression technique 0) and set up some testing on your test environment with DCDATACP N/Y and DCCOMPTC 0/2 for comparison. For example, run a Backup to DASD with different settings:
STEP01 without compression: DCDATACPN
STEP02 with default compression method 0: DCDATACPY + DCCOMPTC0
STEP03 with compression methode 2: DCDATACPY + DCCOMPTC2
Then compare the job step statistics and an ISPF 3.4 listing of the created DASD Archives.
Our performance tests showed that the DASD Archive size with compression technique 2 in relation to the Archive size without compression was 54 %, while it used only little more CPU as without compression. The DASD Archive size with the default compression technique 0 in relation to the Archive size without compression was 44 %, but needed much more CPU time to achieve this.
3) Make sure APAR TR86193 or PTF RO86193 for problem 12413 is applied
This corrects a performance problem when using DCDATACP = Y and DCCOMPTC = 0 on a z/Architecture processor.
CA Disk uses this sysparm in order to determine when data set space usage should be calculated. The default value is 005, which equates to 5 megabytes. Data Sets below this size will not have their IDRC compression rate or tape utilization calculated. When CARTCALC is specified as Y and IDRC compression is used (see sysparm "IDRCUSED"), CA Disk issues I/O requests to tape controllers tracking the compression rate of data written onto cartridges.
On devices attached to 3494 controllers, the overhead generated by the I/O requests is about half a second per data set (1.5 hours for 10,000 data sets). This sysparm allows you to reduce the overhead by instructing CA Disk to only track IRDC compression rate for those data sets larger than the value specified. While this will improve tape I/O performance, it also causes CA Disk to assume zero compression for data sets smaller than the value specified (that is, reduction in output tape usage).