Best practices for Greenplum backups

For the best backup performance all jobs need to run at the same time. For a list of configuration options, see the following topic:

You may be able to improve performance by making the following configuration changes. Make the changes in the following order:

Start with the gpbackup option --jobs 1 and the nbgpbackup configuration option number_of_streams equal to the number of segments per host.
- The value for Maximum jobs per client (host property) should be set to at least the number_of_streams.
- The value for Maximum concurrent jobs (storage unit) should be set as follows:
  (number_of_streams * number of segments hosts) + 1
If the performance is not adequate, then increase the buffer_size or the number_of_buffers.
If the performance still is not adequate, then you can try the following configuration:
--jobs = n
number_of_streams = n * number of segments per host
These settings increase the number of backup jobs that gpbackup starts in parallel.
- The Maximum jobs per client should be at least number_of_streams.
- The Maximum concurrent jobs on the storage unit should be:
  (number_of_streams * number of segment hosts) + 1
If the number of backup jobs is more than the value for concurrent storage unit jobs in the previous bullet, some segment hosts may perform the backup slower than others. If you increase the end_image_timeout value it may keep all segment files in a single image.
If you use disk storage units, you may want to disable throttling. See the following article:
https://support.cohesity.com/s/article/article-100005434