Mainframe, MVS and zOS Discussion

Hi,

To a certain level I understand that VSAM data sets can undergo CI and CA splits as they are updated and records are added or inserted. I have been told that this is not always a bad thing, however sometimes for some data set it can cause excessive processing and I/O. How do we identify such VSAM data sets from the large number of the VSAM we have on our mainframes.

sometimes for some data set it can cause excessive processing and I/O

This is not as true as it used to be.  When a 3390 was a physical box, then yes CI and CA splits could cause issues since they would alter the sequence of sequential reads by causing a jump to a different cylinder then a jump back.  These days with emulated 3390 disk drive, I'm not sure that concern is still valid.  For random processing, it never has been much of a factor anyway since random processing uses the index to jump to the record directly.
If you think this is a concern, there are products on the market to help -- see if your site has one installed.  If your site does not, then your only real option is to generate an IDCAMS LISTCAT on the VSAM data sets and analyze them --manually or by writing a program in the language of your choice to parse the LISTCAT output and look at the CI/CA splits.  There may be other ways to get the data but there still has to be analysis done to find the split counts.

Robert Sample wrote:This is not as true as it used to be.  When a 3390 was a physical box, then yes CI and CA splits could cause issues since they would alter the sequence of sequential reads by causing a jump to a different cylinder then a jump back.  These days with emulated 3390 disk drive, I'm not sure that concern is still valid.  For random processing, it never has been much of a factor anyway since random processing uses the index to jump to the record directly.
If you think this is a concern, there are products on the market to help -- see if your site has one installed.  If your site does not, then your only real option is to generate an IDCAMS LISTCAT on the VSAM data sets and analyze them --manually or by writing a program in the language of your choice to parse the LISTCAT output and look at the CI/CA splits.  There may be other ways to get the data but there still has to be analysis done to find the split counts.

What is emulated 3390? You mean to say that there are no actual device 3390 now and all of the all emulated now? But how do they emulate?

What are the products which can tell about CI/CA splits. If you give some hints I shall search around them.
Thanks for your replies, they are helpful.

[font=Tahoma, Arial, Helvetica, sans-serif]What is emulated 3390? You mean to say that there are no actual device 3390 now and all of the all emulated now? But how do they emulate?[/font]

Once upon a time You could look at a <quite big box> and say ... 
that' s a 3390 disk and data was read and written from large spinning plates
now the technology has changed and the 3390 disk structure is being emulated by the disk controller and the data is spread around on gazillions of smaller disk drives with a different architecture
( the data that was once written on a track is just spread around on different frames of different physical disk drives )

while CI/CA splits are probably irrelevant from a hardware point of view 
VSAM will still incur on some additional processing when doing them

If you really want to know about emulated 3390, find a Redbook on the DS8000 or DS8800 storage system (other vendors such as EMC and StorageTek -- now Oracle -- have similar systems) and start reading.  It's been quite a few years since a 3390 disk drive was an actual physical box behind a 3990 controller attached to the mainframe.
Google vsam tools for more information about various vendor's VSAM products.  As enrico said, the traditional reason for worrying about CI/CA splits -- hardware performance -- isn't really much of an issue any more, but they do incur performance hits in the software as VSAM moves data from CI to CI and CA to CA.  A large number of CA splits, in particular, could impact run times for batch jobs as well as hurt CICS transaction performance.

Thanks enrico and Robert.