DDR LZCOMPACT Option
A new LZCOMPACT option can now be specified on the output I/O definition control statement when using DDR to dump to tape. This provides an alternative to the compression algorithm used by the existing COMPACT option. Unlike the COMPACT option, the data compression done when the LZCOMPACT option is specified can make use of the hardware compression facility to greatly reduce the amount of processing required for compression (DUMP) and decompression (RESTORE). This section summarizes the results of a performance evaluation of the DDR LZCOMPACT option.
Two separate system configurations were used to collect DDR
measurement data. The first configuration consisted of a 2064-109 with
2 dedicated processors, 1G central storage, and 2G expanded storage. The
second configuration consisted of a 9672-R86 with 8 shared processors,
2G central storage, and 1G expanded storage. Two different tape drives
were used: 3590 and 3480 with autoloaders. A typical VM system
residence volume on 3390 was used for the dumps and restores. Multiple
measurements were run in each environment to verify repeatability. CP
QUERY TIME and CP INDICATE USER data were collected for each
measurement.
Table 1. DDR Dump to 3590 Tape: 9672-R86 and 2064-109
Processor Model Compression Type | 9672-R86 NONE | 9672-R86 COMPACT | 9672-R86 LZCOMP | 2064-109 NONE | 2064-109 COMPACT | 2064-109 LZCOMP |
---|---|---|---|---|---|---|
Elapsed Time Total CPU Time CP CPU Time Virtual CPU Time Virtual I/Os Compression Ratio Tapes Required | 1299 9.81 9.26 0.55 100185 1.00 1 | 1256 103.85 8.07 95.78 100185 2.29 1 | 1352 204.98 8.26 196.72 100185 2.50 1 | 1291 3.73 3.56 0.16 100185 1.00 1 | 1253 48.38 3.00 45.38 100185 2.29 1 | 1174 34.11 2.92 31.19 100185 2.50 1 |
Note: z/VM 4.2.0; all times are in seconds |
The LZCOMPACT option reduced elapsed time in the 2064-108 case by 6% as compared to using the COMPACT option. A reduction of 29% was also observed for total CPU time. This was due to hardware compression on the 2064-109. By contrast, CPU time increased on the 9672-R86, which does not have hardware compression.
Another benefit of the LZCOMPACT option was a 9% improvement in the data compression ratio, which reduces tape requirements. We observed an average 10% saving in tape length used per volume when using LZCOMPACT as compared to the COMPACT option, based on DDR DUMP results for a sample of 8 different DASD volumes. The savings ranged from 1% to 28%.
Note that use of the DDR compression options did not affect the
number of I/Os issued by DDR. This is the reason why DDR compression
has rather small effects on elapsed time. The amount of data
transferred per I/O decreases, but this reduces I/O time only slightly
because much of the delay per I/O is independent of the amount of data
being transferred.
Table 2. DDR Restore from 3590/3480: 2064-109
Tape Type Compression Type | 3590 NONE | 3590 COMPACT | 3590 LZCOMP | 3480 NONE | 3480 COMPACT | 3480 LZCOMP |
---|---|---|---|---|---|---|
Elapsed Time Total CPU Time CP CPU Time Virtual CPU Time Virtual I/Os Tapes Required | 988 4.71 4.51 0.20 147659 1 | 996 55.99 3.79 52.20 100828 1 | 981 10.75 3.79 6.96 101364 1 | 1975 4.67 4.47 0.20 147691 11 | 1194 56.04 3.84 52.20 100836 3 | 1169 10.86 3.84 7.02 101372 3 |
Note: z/VM 4.2.0; all times are in seconds |
Use of the LZCOMPACT option reduced total CPU time required to do the DDR restore by 81% relative to use of the COMPACT option, due to the use of hardware decryption on the 2064-109 processor. In the 3590 case, this had no appreciable effect on elapsed time because the restore was I/O-bound and there were no unload/rewind delays since only one tape was involved. In the 3480 case, elapsed time with either compression option reduced elapsed time substantially relative to the no compression case because there were fewer tapes to unload and rewind.