Hub/Switch Installation Guide

Chapter 2 HPSS Planning
HPSS Installation Guide September 2002 67
Release 4.5, Revision 2
Thereare twodifferenttapemigration algorithms,tape volumemigration andtape filemigration. The
algorithm which is applied to a tape storage class is selected in the migration policy for that class.
The purpose of tape volume migration is to move data stored in a tape storage class either
downward to the next level of the storage hierarchy (migration) or to another tape volume within
thesamestorageclass(lateralmove)inordertoemptytapevolumesandallowthemtobereclaimed.
Unlike disk migration, the data is purged from the source volume as soon as it is copied. Tape
volume migration operates on storage segments rather than files. A file may contain one or more
segments. In order for a segment to be a candidate for tape volume migration it must reside on a
virtual volume whose storage map is in the EOM state. Tape volume migration functions by
selecting EOM volumes and moving or migrating the segments off of these volumes. When a
volume is selected for tape volume migration, MPS repeatedly processes lists of up to 3200
segments on that volume until the volume is empty. Once all of the segments have been removed
from a volume, that volume automatically moves into the EMPTY state and may be reclaimed for
reuse. MPS continuesthisprocessuntileither thepercentageofvolumes specifiedinthe migration
policy are emptied or no more EOM volumes can be found. Segments on an EOM volume are
evaluated for tape volume migration based on the values in the migration policy for that storage
class. If asegment hasbeeninactive for a sufficientlengthoftime it will be migrated. If asegment
has been active within the configured amount of time, or if any other segment in the selected
segment's file has been active, theselected segmentwill bemovedlaterally. The Migrate Volumes
and Whole Files option in the migration policy allows all of the segments belonging to a file to be
migrated together, includingthose segmentswhich resideonother, potentiallynon-EOM volumes
than the EOM volume which is being processed. This option tends to keep all of the segments
belonging to agiven file at the samelevel in the hierarchy. Ifa segment is selected formigration by
MPS, then all other segments belonging to the same file, regardless of their location, will be
migrated during the same migration run. If any of the segments in the file are active then none of
them, including the segment on the selected EOMvolume, willbe allowedto migrate. Rather, the
selected segmentwill be movedlaterally and none of theadditional segments willbe moved atall.
Tape file migration can be thought of as a hybrid between the disk and tape volume migration
algorithms. Disk migration is a file based algorithm which is strictly concerned with making one
ormorecopies ofdisk files.Tapevolume migrationis onlyconcerned withfreeingtape volumesby
movingdatasegmentsfromsparselyfilledvolumeseitherlaterallyorvertically.Tapefilemigration
is a file-based tape algorithm which is able to make a single copy of tape files to the immediately
lower level in the hierarchy. Similarly to disk migration, tape copies are made roughly in file
creation order, but the order is optimized to limit the number of tape mounts.
As with disk files, BFS creates migration records for tape files in storage classes which are using
tape file migration. MPS reads these migration records in the same manner as it does for disk.
Within a given storage class, files are always migrated in hierarchy and family order. Hierarchies
arecheckpointed in thesame way asdisk. Files are roughly migrated bycreation time(the time at
which the first file write completed), but priority is given to migrating all of the files off of the
current source volume over migrating files in time order.
When a tape filemigration run begins, all ofthe eligiblemigration records for the storage class are
read. Foreach migrationrecord,the tapevolumecontaining thecorrespondingfileis identified. A
list of all of the source tape volumes which are to be migrated in the current run is created. MPS
then begins creating threads to perform the actual file migrations. A thread is created for each
source volume up to the limit specified by Request Count in the migration policy. These threads
then read the migration records corresponding to their assigned volumes and migrate each file.
The migration threads end when the supply of migration records is exhausted. As each thread
ends,MPS startsanother threadforthenext sourcetapevolume tobe migrated. Themigrationrun
ends when all volumes in all hierarchies have been migrated.