In backup runs, packet generation depends on the type of the file being processed:
BS2000 files
In backup runs, division into packets takes place in accordance with the catalog ID and the user ID.
In the case of PARALLEL-RUNS=1, there is one packet per catalog ID and user ID. However, as soon as PARALLEL-RUNS>1 is specified, HSMS divides the files of a catalog ID and user ID into four packets according to their location on the pubset disks.
When the number of pubset disks is greater than three, all four packets contain files. When the number of pubset disks is less than three, the number of filled packets is the same as the number of pubset disks. This means that the other packets of the four remain empty.
Packet division thus also permits parallel processing when files are saved from just one catalog ID and user ID. Division into smaller packets also results in more even utilization of the parallel subtasks, and therefore reduced backup times.
Division as described in section “User control of packet generation” is only still to be recommended in exceptional cases, e.g. when saving very few but very large files.
Files on Net-Storage
For the files on Net-Storage, HSMS generates one packet per catalog ID and user ID for each Net-Storage concerned irrespective of the specification in the PARALLEL-RUNS operand.
Node files of passive nodes S0
In save runs, node files from various passive nodes are assigned to separate packets. First the specified number of ARCHIVE subtasks is started and subsequently packets created. Initially the lowest common directory level is determined from the specified node files in a node and then mutually independent subtrees are determined at this level. If there are less free subtasks than subtrees, a number of subtrees are combined. In each subtree a packet is created for each directory level; all packets are allocated to a subtreee, and with this all files from a file tree processed by the same subtask.
This structuring applies both for PARALLEL-RUNS and MULTIPLEXING.
If the file specification *ALL is being used in a save run for node files in BS2000 so that the include/exclude files are evaluated on the associated client, internal structuring into subtrees takes place beforehand as if the Include/Exclude specifications had been made in BS2000.
With RESTORE and COPY runs the structuring into packets depends on how the data was saved to the volumes.