| Parameter | Description |
|---|---|
| Parallelism | Number of channels RMAN allocates for backup and restore operations (typically aligned with available CPU cores and I/O capacity) Sizing guidance: RMAN channel count and section size should be sized together based on available CPU resources, database size, and the number and size of datafiles. The goal is to provide enough parallel work to keep channels busy without overwhelming CPU or increasing file-management overhead. |
| Section Size | Splits individual datafiles into chunks for parallel processing across multiple channels |
| Files Per Set | Maximum number of data files bundled into each backup set |
| Max Piece Size | Maximum size of each backup piece file (applies to backup sets only, not image copies) |
RMAN parallelism is driven by three factors:
-
Multiple datafiles that can be processed concurrently
-
Sections within large datafiles that allow multiple channels to work on a single file
-
Multiple channels to process the work
The key parameters control how work gets divided, so channels stay busy:
-
SECTION SIZE splits individual datafiles into chunks that different channels can process in parallel. Without SECTION SIZE, RMAN assigns one channel per datafile — a 500GB bigfile tablespace would be processed by a single channel while others sit idle.
-
With SECTION SIZE 4G, RMAN splits that 500GB file into approximately 125 sections. Multiple channels, whether on one node or distributed across RAC nodes, can process those sections in parallel.
-
There is a cost associated with opening and closing sections; setting SECTION SIZE too small increases file management overhead and can reduce backup efficiency. In practice, SECTION SIZE values of 16GB or larger are commonly recommended.
-
Oracle limits the maximum number of sections per datafile to 256, so SECTION SIZE should not be set smaller than approximately (datafile size ÷ 256). SECTION SIZE is especially important for large datafiles, bigfile tablespaces, and multi-node RAC operations where channels across instances need sufficient work to stay busy.
-
-
FILESPERSET controls the maximum number of datafiles bundled into each backup set. Lower values create more backup sets, providing more units of work for channels. With FILESPERSET 1, RMAN creates a separate backup set for each datafile. With FILESPERSET 8, up to eight datafiles bundled together. When SECTION SIZE is used for large files, FILESPERSET becomes less critical, since parallelism is driven primarily by sections within datafiles rather than the number of files per backup set.