IBM AIX - Logical Volume Manager (LVM)
Concepts:
• LVM (Logical Volume Manager) is the ever-present
disk and volume management framework for AIX. The level of integration
is visible not only in fileystem commands that understand the underlying
LVM, but in other, higher level, commands like the install and backup
utilities that can optionally grow filesytems when necessary.
• Physical disks (hdisks) are placed under LVM
control by adding them to a VG (volume group). Within LVM, these disks
are referred to as PVs (Physical Volumes).
• Each PV in a VG contains a unique ID called a
PVID. The PVID of a disk is used to track all disks in a VG, but also
provides a device name independence that makes importing, exporting, and
disk management much simpler. Because the unique characteristics of the
disk become the identifier, the device name remains consistent but does
not need to as (properly) renaming / reordering disks under LVM control
is of little consequence.
• Once a hdisk is placed into a VG it is divided
into PP (Physical Partitions). PPs are then used to create LVs (Logical
Volumes). An additional layer of abstraction is placed between an LV and
a PP called a LP (Logical Partition) that allows for more than one PP
to be used (i.e. mirrored) to back each portion of a LV.
Several on-disk structures are responsible for
holding all LVM information. The VGDA resides on each disk and holds
structural information such as the member PVs. The VGSA also resides on
each disk and contains status information on all member devices. The
LVCB varies by VG type but traditionally has resided in the first part
of an LV (when it exists as a separate structure). In addition to the
basic LVM commands that manage these structures, there are a number of
lower level LVM commands that accesses this metadata more directly.
• The first disk in a VG will have two copies of
the VGDA, and a two disk VG will have one disk with a single VGDA and
the other with two copies. For three disk and larger VGs, each disk has a
single copy of the VGDA.
• The concept of quorum is achieved when > 50% of the copies of the VGSA/VGDAs are online. If quorum is lost then the VG can be taken offline.
• Quorum is problematic for two disk VGs because
the loss of the two VGDA disk means a loss of the entire VG. In a
mirrored configuration (a typical case for two-disk VGs) it is
inappropriate to offline the VG for a single disk failure. For this
reason, quorum rules can be turned off in the case of a two disk
mirrored VG.
The ODM is central to managing off-disk LVM
structures and physical device to hdisk mappings. When a VG is created
or imported this information is added to the ODM as well as other system
files such as
/etc/filesystems.
• AIX LVM supports several versions of VGs that
have been introduced over the lifetime of the product. The VG types are
normal, big, and scalable. Normal VGs were the original creation and are
more limited than the big or scalable types. The easiest way to tell
the type of an existing VG is to look at the Max PV value for the VG
(see example in the next section).
The default filesystem on AIX is JFS2. JFS2, and it
predecessor JFS, are both journaling filesystems that utilize the
fundamental Unix filesystem structures such as i-nodes, directory
structures, and block allocations. (Technically, JFS2 allocates blocks
in groups called "extents".)
• JFS2 is not an implementation of UFS and expands
considerably over basic filesystem features with such capabilities as
snapshots, dynamic i-node allocation, online growth, extended
attributes, and encryption. AIX provides a layer of abstraction over all
supported filesystems that map filesystem specific structures to
standard Unix filesystem tools so that filesystems like JFS2 appear as
an implementation of UFS.
• While most journaled Unix filesystem
implementations use inline logs (within the filesystem structure), AIX
tends to use a special type of LV that is created only to contain log
data. The jfs(2)log LV can provide logging capability for more than one
filesystem LV. The log type must match the filesystem type. JFS2 can log
to an inline log, but these implementations tend to be the exception to
the rule.
