KSDS sizing: picking CI/CA for ~337M records — rules of thumb?

[Abhishek Awasthi]

We're defining a high-volume KSDS for ~297M rows, avg record 449 bytes.
Considering CI=32768 and CA=65536 for KSDS; data volume ~163M records.

Any recent guidance to balance insert performance, freespace, and CA splits? IDCAMS DEFINE examples welcome.

 

[Robert Sample]

It definitely looks like you don't know what you're talking about -- your post title mentions 337M records while your post says 297M rows and then there is "data volume 163M records".  I'm assuming you mean millions for each M although you did not specify.  Questions you need to research include:
- What is the insert frequency?  Whether there is one insert per day, or 10 million inserts per day, can change recommendations.
- Are the inserts random throughout the data set or do they only occur in certain spots?
- Is the KSDS going to be used in CICS or batch only?
- Are the records fixed length or variable?  If variable, what is the minimum and maximum lengths?
- For fixed length 449 byte records, it appears that 18432 would be a slightly better CI size than 32768.
- How often will the dataset be unloaded and reloaded to clear up splits?
- For relatively few inserts, you might want to set FREESPACE to (0,0) and take the performance hit of CA splits when inserting records.
- For relatively many inserts, you will want to set your CI FREESPACE value high enough to allow at least one record to be inserted.