Chapter 7 HPSS User Interface Configuration
444 September 2002 HPSS Installation Guide
Release 4.5, Revision 2

7.6.2.8.1 Migration/Purge Algorithms and the MPQueue

Migrationand purge are handled differently in an HPSS/XFS system than in an HPSS/DFS system.
In an HPSS/XFS system, a queue of migration and purge candidates are kept in shared memory
usingthe Migration/Purge Queue or MPQueue. When migration and purge run, they simply step
through the migration and purge candidates in the MPQueue. This eliminates the need to query
every file in the filesystem for each migration and purge run and will result in much more
expedient migration and purge runs than for a comparable DFS filesystem.
However, since the MPQueue resides in shared memory it must have a fixed size. This size is
decided when the filesystem is initially configured and should be planned carefully, taking into
consideration the migration and purge policies to be used and expected filesystem activity. It will
benecessary to have one MPQueue entry available for each unmigrated file as well as each file that
has been migrated but not yet purged. In other words, any file which is not completely migrated
andpurged from XFS will take up an entry in the MPQueue. It is recommended to use an MPQueue
size that is at least twice the maximum number of “active” files expected for the file system.
7.6.2.9 Other Limitations
These limitations apply equally to HPSS/DFS and HPSS/XFS systems.

7.6.2.9.1 HPSS/DFS Activity

Thesystem administrator must consider what level of HDM/DMG activity the core HPSS servers
cansupport. HPSS speed is limited by hardware, SFS, disk and tape speeds. If HDM/DMG activity
causes millions of HPSS file creations and file migrations into HPSS, then the system may have
problems.
Users of managed filesystems should be educated about the limitations of such configurations.
Thoughthe HDM/DMG interface allows for normal Unix type file activity, it can not be forgotten
thatthese interfaces access data from the archive. The HDM is implemented to migrate individual
files into HPSS, so there is a potential that each file access may incur a tape delay when the file is
readback from HPSS. If an HDM/DMG user stores related data into larger files, the total number
of tape accesses can be decreased, increasing the user’s performance. This will also decrease the
load on HPSS.
Usersand administrators should be educated in the use of managed filesystems, so that informed
decisions can be made about their use. A given fileset type does not optimally handle all types of
fileusage patterns. Standard DFS and XFS filesystems run out of disk space. Archived filesets incur
delays for data retrieval if the data has to be staged back to the managed filesystem. HPSS/DFS
mirrored filesets incur delays for data retrieval and also are slower for name space updates, but
they permit access to the data and name spaces through all HPSS interfaces.
7.6.3 Configuration
For AIX systems, it is assumed that the system on which the DFS Server is running has been
configuredwith the appropriate DCE and DFS versions and PTFs as given in Section 2.3.2.1: HPSS
Server/Mover Machine - AIX on page 50.