FASTCPK
Overview | Features | Benefits
| Case Studies | Brochures
A recent article in Enterprise Systems Journal ("The Case Against DEFRAGs," July 1995) described the deficiencies of IBM's DFDSS (or DFSMSdss) DEFRAG in today's 24x7 data center. The article described in detail the problems with DEFRAG, such as long run times and high I/O overhead. It also questioned the need for the simple objective of DEFRAG, namely the collection of fragmented freespace into fewer areas of larger size to avoid the possibility of allocation failures due to insufficiently large free areas.
The author's solution involves the use of systems-managed storage to direct allocations to various pools of disk volumes based on the size of the data set. The volumes containing smaller data sets will become more fragmented, but they have smaller primary allocations and so can tolerate more fragmentation.
We can't argue with the author's analysis or his conclusions in the right environment, but there are some other factors, especially if Innovation's COMPAKTOR product is used instead of DFDSS DEFRAG.
Although systems-managed storage has been available for six years, there are still many data centers that have not implemented it and others that still have many non-systems-managed storage volumes. According to recent industry surveys, less than 50 percent of the volumes in MVS shops today are systems-managed storage. So the suggested environment for the avoidance of DEFRAG is often not practical.
The author downplays the impact of multiextent data sets, and it is true that with today's higher-speed disks, cached control units and RAID implementations, the performance degradation of multiple extents is reduced. But it is still true that many data sets have a maximum of 16 extents; even on systems-managed storage volumes, only the newer PDSE and Extended Format data sets (plus VSAM, of course) can exceed 16. Underestimation of the space required for a data set may lead to X37 abends as the maximum extents are exceeded.
Innovation Data Processing has been addressing the volume reorganization concerns of our customers since 1975, long before DEFRAG was introduced. For more than 20 years, we have provided a tool, COMPAKTOR, that will not only collect freespace into larger areas but also can merge multiextent data sets into single extents and release overallocated space; these functions include PDSEs and VSAM. The majority of MVS data centers, over 5000, are licensed for COMPAKTOR. In June 1993, we introduced the FASTCPK option of COMPAKTOR which provided significant performance enhancements. The average time to COMPAKT a 3390-3 volume, while reducing the number of freespace areas up to 90 percent, is less than four minutes.
The results from a typical large user are shown in the accompanying chart. This actual user COMPAKTed 391 DASD volumes, including a mixture of 3380, 3390-2 and 3390-3. These volumes are separated into three pools: a large data set pool with fewer than 100 data sets per volume, a medium pool with about 300 data sets per volume, and a small pool with over 1000 data sets per volume. You can see that the user had already implemented the solution suggested in the article, yet felt the need to COMPAKT the volumes.
Some options that worked well with the old COMPAKTOR and older DASD technology can be detrimental to FASTCPK. Many customers previously used COMPAKTOR to position data sets to reduce head movement. As the article correctly points out, this is now of minimum benefit and can actually increase FASTCPK run time.
The recommended control statement for the fastest COMPAKTion with the least movement of data while still gaining most of the benefits of COMPAKTion is:
CPK TYPE=FASTCPK,VOL=vvvvvv,SIZEKEEP=(100,90,60),CPKFREEX=10
This instructs COMPAKTOR to achieve a 90-percent reduction in the number of freespace areas and only merge those multiextent data sets needed to reach a 90-percent freespace reduction.
AVG AVG
# of # of Free Area % Reduction MultiExtent DSN Elapsed Time
VOL Type DSN Before After In Free Area Before After (Min.)
10 3380-E 1061 4067 20 99% 363 20 5.6
8 3380-K 1145 3387 16 99% 298 5 5.4
3 3380-E 1051 1165 6 99% 96 0 4.7
2 3390-2 1127 892 3 99% 78 0 4.4
10 3380-E 208 860 19 97% 91 0 5.7
4 3390-2 258 95 2 97% 14 0 4.7
10 3390-3 71 336 35 89% 44 32 3.2
4 3390-3 67 114 11 90% 18 12 2.7
10 3390-3 71 353 39 89% 35 24 1.9
10 3390-3 76 340 33 90% 47 37 3.6
In the small and medium pools, the COMPAKTOR options were set to reduce the number of freespace areas to the minimum (usually one or two), while the large pool was set to reduce the number of free areas by up to 90 percent.
The result was an average elapsed time per volume of 4.5 minutes with 96-percent reduction in freespace areas (from 32254 areas to only 1314 areas across all the volumes). Multiextent data sets were reduced by 76 percent (3357 such data sets before and only 785 after). In addition, overallocated space was released on sequential (PS) data sets during the COMPAKTion, resulting in 201,533 tracks being added to the freespace.
From extensive analysis of real COMPAKTOR runs at customer sites and tests at our own data center, we have seen comparable results in most cases, even when 20 COMPAKTions are run in parallel.
Our experience on EMC (Hopkinton, MA) disks and early experience on RAMAC (IBM Corp.) and ICEBERG (Storage Technology, Louisville, CO) have shown even better results, since these new devices have even faster access times and make more efficient use of cache.
Before Innovation introduced the FASTCPK option of COMPAKTOR, as DASD devices grew larger there was a significant migration to DFDSS DEFRAG since the older COMPAKTOR required a full-volume backup as part of the process, so COMPAKTOR's elapsed time kept increasing (up to 35 minutes for a 3390-3). As you can see, with FASTCPK, the average elapsed time is only a few minutes. Not only have many former COMPAKTOR users switched back since FASTCPK, but a significant number of DEFRAG users have switched to FASTCPK.
In summary, the efficiency of FASTCPK compared to DEFRAG and the added value of the additional functions of COMPAKTOR make it practical and useful to do disk reorganizations.
Bruce Black
Senior Software Developer
Innovation Data Processing
Little Falls, NJ
Reprinted by permission of ENTERPRISE SYSTEMS JOURNAL, P.O. Box 740908, Dallas, TX 75374-0908, (214) 669-9000.
All rights reserved. Free subscription information available upon request.
|
