Q and A

Asked and Answered

Does DASD striping work with DB2 for z/OS and OS/390? After discussions with IBM reps, my company decided to use it in some of our large flat files. IBM said it worked with VSAM and that it should work with DB2. I thought it would have a big benefit with the load and unload utilities and cursors that pull in large volumes of data. What are your thoughts?

Robert Catterall responds:

That's an interesting question. In my experience, DASD striping is usually mentioned in discussion of storage management for DB2 in a Unix, Linux, or Windows environment. I rarely hear the term used when the operating system DB2 is running under is OS/390 or z/OS. I'm not a storage systems expert, but my hunch is that striping isn't a big deal in DB2 for z/OS and OS/390 shops because it's not really needed. Yes, disk subsystem throughput and response time is very important in a mainframe DB2 environment, but outstanding performance can be achieved without disk striping.

Why? Well, the very large amount of cache storage on today's high-end disk controllers (often multiple gigabytes per control unit) is one important factor. With a high percentage of cache hits, you don't go so often to spinning disk in a synchronous way. Furthermore, DB2 for OS/390 and z/OS databases are often very large (at least hundreds of gigabytes and often several terabytes). That means lots of disk volumes. Lots of volumes means the odds of a good spread of high-use data sets are good (fewer "hot spots" in the disk subsystem). Then there's tablespace partitioning, which means something quite different for a mainframe DB2 system versus the term's meaning in a DB2 for open systems environment. It's common practice for DB2 for z/OS and OS/390 DBAs to partition a large table, spreading the table rows (and the entries in the partitioning and clustering index) across up to 254 data sets, which, given a large pool of volumes, should spread I/O activity out effectively.

Finally, high-end RAID disk subsystems achieve a degree of data striping under the covers (at least that's my understanding).

Bottom line: I suppose you could work on disk striping in a mainframe DB2 environment, but would your investment of time and effort pay off? At my company, we don't worry about striping, and synchronous read times for our production DB2 for OS/390 system are typically 5 milliseconds or less.

See a complete archive of reader/author Q&As .