User's Manual

126 Migration, release, recall, and deletion
job requires new resources. Typically, the Resource Manager locks the use of resources in the Resource
Management Database for jobs that have requested allocation.
Media selection by media location
Allocation selects a medium according to its location. Possible locations of a medium are:
• Online: Medium is loaded into a drive.
• Nearline: Medium is in a library slot.
• Offline: Medium location is unknown.
The Resource Manager always allocates media and drives that are in the same library first to avoid
time-intensive load/unload operations. If there are still jobs in the queue and no online media are
available, then nearline media are selected.
Whenever possible, the current volume from an allocated medium is used. Otherwise, the volume with the
highest use is selected to minimize the number of open volumes.
Several allocation parameters
There is only one Resource Manager for the whole FSE implementation, and therefore it is responsible for
all FSE partitions. The parameters needed for resource allocation are different for the whole FSE
implementation (system) and each partition; they are defined in the systems policies and in the partitions
policies.
Note that resource allocation requests from different partitions can have different drive and priority
policies, and that the same media pool cannot be assigned to several partitions. For more information on
parameters for resource allocation, see ”System allocation and job priority policy” on page 127 and
”Partition allocation and job priority policy” on page 128.
Job priorities
When a new job is created, if all of the requested resources are available in the FSE implementation,
Resource Manager allocates them and the job starts performing its task immediately. If some resources are
not available, a resource request of the job is put into the resource allocation queue of the Resource
Manager, and the job waits for availability of the resources.
In case with too many jobs already running for the corresponding FSE partition, the newly created job is
not allowed to request resources yet, but is put into the state ”pending” instead. Such pending job waits
until the current job count for the FSE partition decreases below a threshold value, and can only request the
resources from the Resource Manager afterwards. Pending jobs are processed in the “first come first
served” order. The threshold job count is defined by the MaxNumDrives variable (specified in the FSE
partition configuration file) as follows:
IMPORTANT: The status “pending” is applicable only for jobs that can be run on a particular FSE
partition: migration, recall, reorganization, and maintenance jobs. Administration and backup jobs are
FSE implementation-wide jobs. They are immediately started or queued in the resource allocation queue,
no matter how many jobs are already running in the FSE implementation.
When the resource request of a job is put into the resource allocation queue, the job is assigned an initial
priority. When a new job is put into the queue, the priorities of all existing resource requests are
recalculated; the priority of each job is then modified according to the elapsed time and the phase of the
job. The resource request is then processed according to its job priority. Each time a resource is freed, the
queue is checked again to allocate the free resource.
IMPORTANT: Only jobs with the status “waiting for resources” have priorities assigned.
For details on priority calculation, see ”Priority calculation” on page 130.
ThresholdJobCount = 3/2 * MaxNumDrives