VSAM RRDS: Relative Record Data Sets

Every record in an RRDS occupies a specific slot number (1, 2, 3, ...), and access is always by that numeric position — there's no separate index component like a KSDS, because the relative record number itself directly maps to a physical location within the dataset, making RRDS access extremely fast and predictable, arguably faster than KSDS for pure direct access since there's no index traversal at all. The tradeoff is that the 'key' has no inherent business meaning; the application itself must maintain the mapping between a meaningful identifier and the correct relative record number.

RRDS comes in two flavors: fixed-length RRDS, where every slot is the same predetermined size (ideal for tables of records with a natural sequential numbering, like a fixed set of 366 day-of-year records or a bounded set of parameter/configuration rows), and variable-length RRDS (VRRDS), which relaxes the fixed-size constraint at some cost to the simplicity of the pure slot model. Because slots can be genuinely empty (never written) as well as occupied, RRDS also naturally supports sparse population where certain relative record numbers are simply never used.

A sharp interview answer situates RRDS honestly: it's a comparatively rare choice in modern applications specifically because most business data needs meaningful key-based lookup (which is what KSDS provides), and RRDS shines mainly in narrower scenarios — capacity or configuration tables addressed by a naturally numeric position, or specialized high-performance lookup structures where the application already knows or can compute the correct relative record number without needing a general-purpose index.