Managing smf, Co-location of data – HP Integrity NonStop J-Series User Manual
Page 58
Configuring and Managing SMF Processes
HP NonStop Storage Management Foundation User's Guide—523562-007
3-8
Managing SMF
Assume that we have two virtual disks, $V1 and $V2 and two physical disks, $P1 and
$P2 in one SMF pool, $PP1. A file TEST.PART has its primary partition on $V1. Its
secondary partitions are
$V2.TEST.PART
$P1.TEST.PART
Also, assume that the physical file corresponding to the secondary partition
$V2.TEST.PART is $P1.ZYS00000.A00000M2. This means that the secondary
partition $P1.TSET.PART and the physical file corresponding to the secondary partition
$V2.TEST.PART are on the same physical disk, $P1. In case physical disk $P1 fails,
you would lose both of the secondary partitions. You should avoid this situation.
Use the PHYSVOL option to create secondary partition $V2.TEST.PART and place its
physical file on $P2 or use the FUP RELOCATE command to move the physical file
from $P1 to $P2.
Co-Location Of Data
Many users optimize applications by co-locating certain data files on the same disks or
by spreading data across multiple disks for maximum parallelism. Before moving an
application to SMF, you should understand how the application is currently tuned so
that optimization can be preserved. Co-location of data is preserved when a disk is
converted by using SMCONVRT. It can also be preserved when using RESTORE by
specifying the PHYSVOL option, which forces all the restored files to be located on the
same physical volume.
Conversely, if you do not want co-location, you can spread data across the entire pool
with RESTORE by omitting the PHYSVOL option and allowing SMF to choose the file
locations. If you later find that certain files are not located correctly, you can use the
FUP RELOCATE command to move them to a new location within the pool.
When using SMF to migrate data to newer, larger disks, it is important to avoid head
contention. If several files containing frequently accessed data formerly located on
separate disks are all placed on the same large disk, performance degrades. Such files
should be spread across all the disks in the pool. You can also use your knowledge of
data access patterns to distribute the data. Suppose the data is stored by date and you
must produce reports that do year-to-year comparisons. If each month's data is on a
different volume than the data from 12 months earlier, you will avoid head contention.
Managing SMF
All SMF processes run as privileged, multithreaded process pairs. Only one $ZSMS
process pair executes on any node, but multiple storage pool and virtual disk
processes can execute on a system. Each storage pool process manages one storage
pool, and each virtual disk process is associated with a particular storage pool. This
subsection summarizes operations using COUP and PUP and SCF to define and
manage SMF processes and their associated storage pools.