Control areas (CAs) group control intervals so VSAM can manage space in chunks bigger than a single CI. When a CI split needs a free CI and none remain inside the current CA, VSAM performs a CA split: allocate a new CA, move roughly half of the CIs, update the index, then finish the CI split. CA size tuning is therefore about how much data lives in each blast radius and how many free CIs you keep at the CA level via FREESPACE. Beginners often master CISZ first and forget the second FREESPACE number; this page closes that gap with intuitive explanations, trade-offs, and operational habits that keep CA splits rare and predictable.
Picture a bookcase (CA) with shelves (CIs). When one shelf overflows, you slide books to an empty shelf on the same bookcase—that is a CI split. If every shelf is full, you must buy a new bookcase, move half the shelves to it, and only then free shelves on the original bookcase—that is a CA split. CA tuning decides how heavy each bookcase is. Heavy bookcases mean fewer bookcases in the room but a painful move when you finally need a new one. Light bookcases mean moves are smaller but you might walk across more bookcases during a long sequential read.
| Lever | Upside | Downside |
|---|---|---|
| More CIs per CA | Amortizes sequence set entries over more data; can reduce relative index depth growth rate | CA split work moves more bytes when it happens |
| Fewer CIs per CA | Smaller blast radius per CA split | More CAs per gigabyte; more sequence set traffic for wide scans |
FREESPACE( ci-percent ca-percent ) reserves two kinds of emptiness. The first reserves bytes inside each CI for inserts without an immediate split. The second reserves entire empty CIs scattered through each CA so that when a CI does need to split, VSAM can allocate a fresh CI in the same CA instead of escalating to a CA split. Setting the CA percent to zero is a common accidental cause of CA split storms on growing KSDS files. A moderate value such as 10 to 15 is a starting conversation point for insert-heavy files; static archives might use lower values to avoid wasting allocated space.
Long sequential jobs read CIs in ascending address order. Larger CAs mean each CA boundary crossing happens less often, but each physical stage may still be CI-sized depending on buffering. Do not assume CA size replaces BUFND tuning; buffers decide how many CIs stay resident. CA tuning and buffer tuning are complementary chapters in the same performance story.
Document baseline CA geometry, split counters, and business date ranges before any rebuild. After migration, compare the same counters across equivalent calendar windows. Performance tuning without measurement is guesswork; measurement turns arguments into engineering.
CA splits need unused space inside the cluster allocation. If secondary space is exhausted, inserts fail even if theory says a CA split could help. Capacity planning must align CA tuning with CYLINDERS or RECORDS allocation and SMS data class thresholds.
The sequence set is the lowest index layer that points to CIs. When you grow the number of CIs in the data component, the sequence set acquires more entries. CA boundaries influence how many CI descriptors appear in contiguous address ranges on disk, which in turn affects how many index blocks must be touched during a long forward scan. You do not hand-edit sequence set blocks, but you should understand that CA tuning indirectly shapes index locality: extremely small CAs can scatter CIs across more sequence set entries per megabyte of user data, while very large CAs keep more CIs behind each pointer but raise the cost of structural moves. Performance specialists sometimes model this with LISTCAT statistics plus SMF-driven I/O histograms rather than guessing from geometry alone.
SMS data classes may impose minimum allocation or extent rules that interact with how many CAs fit per extent. When you request a geometry change, bring three artifacts: current LISTCAT (cluster and components), a graph of CI and CA splits over the last month, and a short narrative of the business workload (insert-heavy month-end versus static reference). Administrators can translate that into volume placement and extent sizing. If you only say "make CA bigger," you might get a larger number on paper without the FREESPACE CA discipline that actually prevents splits. Partnership beats ping-pong tickets.
The second FREESPACE number is like leaving a few empty shelves on every big bookcase so you can slide books when a row gets too full. If you forget to leave empty shelves, you must carry half the books to a brand new bookcase in the middle of homework time—that is slow and messy.
1. What does the second FREESPACE number reserve?
2. Why is CA split called expensive?
3. Which statement is most responsible?