Performance tuning considerations for Oracle databases with bigfile tablespaces
This applies to:
Exporting an Oracle PDB to a Physical Filesystem-based Container Database
Exporting an Oracle Dataset to a Physical ASM or Exadata Database
As part of the export workflow, Delphix performs the data transfer using RMAN. By default, 8 RMAN channels are created to copy files from Delphix to ASM or Physical Filesystem. The number of RMAN channels can be specified in the export CLI. Each RMAN channel takes up one datafile at a time and picks up another as soon as it is done copying one. For databases that contain hundreds or thousands of small data files, this is usually sufficient.
Consider a database which contains 100 data files of sizes 20-40GB each, and 1 large datafile of size 1TB or larger. In such a case, one of the RMAN channels will be doing all the work of transferring the 1TB datafile and the rest of the RMAN channels would be idling until the 1TB of data is copied. To speed up the data transfer for such databases, Delphix now provides a new rmanFileSectionSizeInGb API parameter that can be passed to the export CLI. When this parameter is specified, Delphix will pass the SECTION SIZE parameter to the RMAN BACKUP AS COPY command, and RMAN then creates several chunks of backup copies in which each chunk contains the blocks from one file section. This type of copy is called a multisection copy. The purpose of multisection copies is to enable RMAN channels to copy a single large file in parallel. Thus, when the rmanFileSectionSizeInGb parameter is specified in the export CLI, RMAN will divide the work among multiple channels, with each channel copying one file section in a file. For example, if rmanFileSectionSizeInGb is set to 64, then the large 1TB datafile transfer will be broken into 16 sections of 64GB each, and all 8 channels will then be utilized for copying the large datafile once the copy of the smaller files has completed. Copying a file in separate sections can thus improve the performance of copies of large data files which in turn could speed up the time taken by the V2P export job.