LVM

LVM Theory

Physical Volume = PV is IBM speak for a disk (it could be worse some IBMers still refer to DASD! (Direct Access Storage Device which is a mainframe term). They are:

• Named by AIX as hdisk0, hdisk1, hdisk2, ...
• Regardless of the underlying technology (SCSI, SSA sort of IBM's early SAN, SAN, RAID5 using the adapter)
• Disk and AIX automatic bad block reallocation

Volume Group = VG is IBM speak for a group of disks
Volume Group operations:

• Disk space always allocated within single VG
• All disks available in AIX or none - work as a group
• Can be exported to be attached to other AIX - allows High Availability HACMP
• First VG called rootvg

Root Volume Group (rootvg) is created automatically while installing AIX is placed within this VG

• AIX files
• Initial paging
• Usually only the first disk
• Or two, to allow mirroring of rootvg
• Often internal disks
• Recommend: Keep to a small number of disks

Other Volume Groups

• Other VG created by the System Admin.
• Name typically related to their use
• Optionally ending in "vg"

Volume Groups

• One or more disks
• Can later add a disk to a VG
• Can remove disk from VG - if empty
• Disks within VG cut into chunks called
• Physical Partitions (PP)
• Minimum allocation unit
• Typically, 8, 16 or 32 MB (older 2GB disks = 4 MB)
• Nothing smaller can be allocated
• Each VG fixed PP size.
• Volume group has total, free and used PPs
  *Disks do not need to be the same size or even the same technology

Logical Volumes

• It is a large piece of disk
• Can be used for lots of tasks like
• raw device or file system or paging space
• Created within single VG
• LV has many attributes like:
  • can be one disk or more disks or all disks of VG
  • mirrored or striped
  • particular part of disk
  • Sizes is in Physical Partitions - example: id PP = 16 MB and LV size of 64 PPs = 1 GB
    PPs allocated by policies
• min or max spread across disks
• min=fill disk then move on, max=evenly spread
• disk placement centre, middle, edge
• mirrors via extra copies
• stripes via stripe width
• MWC (Mirrored write consistancy) and others

Logical Volumes Admin

• Can be grown in size
• Can be removed
• Can be moved to different disks within VG
• Can later have a filesystem created in it
• Attributes can be changed like: mirror added or removed or policy changed

How AIX/LVM sees the underlying disks

LVM in practice

The disks can be apprached from three different directions and from there you can see the relationship to the others.
As a reminder

• pv = physical volumes - real disk or LUN
• lv = logical volume
• vg volume group
  

To the this use the ls<thing> commands:

• lspv
• lslv
• lsvg

In more detail

• Lists volume group names: lsvg
  • List VG details: lsvg <vgname>
  • List LVs within VG: lsvg -l <vgname>
  • List PV within VG: lsvg -p <vgname>

• Lists physical group names: lspv
  • List PV details: lspv hdiskN
  • List LVs within PV: lspv -l hdiskN
  • List PP within PV: lspv -p hdiskN

• Lists LV names: lsvg -l <vgname>
  • List LV details: lslv <lvname>
  • List LVs within PV: lslv -l <lvname>
  • List PP within PV: lslv -p <pvname>
  • Show mirrored LV: lslv -m <lvname>

When you create a Logical Volume with smitty you will see a panel like this:

smitty create logical partition

Add a Logical Volume

Logical volume NAME                      [scratch]

  VOLUME GROUP name                         testvg

  Number of LOGICAL PARTITIONS             [64]                     #

  PHYSICAL VOLUME names                    [hdisk1 hdisk2 hdisk3]   +

  Logical volume TYPE                      [jfs]

  POSITION on physical volume               middle                  +

  RANGE of physical volumes                 minimum                 +

  MAXIMUM NUMBER of PHYSICAL VOLUMES       []                        #

    to use for allocation

  Number of COPIES of each logical         2                       +

    partition

  Mirror Write Consistency?                yes                     +

  Allocate each logical partition copy     yes                     +

    on a SEPARATE physical volume?

  RELOCATE the logical volume during       yes                     +

    reorganization?

  Logical volume LABEL                     []

  MAXIMUM NUMBER of LOGICAL PARTITIONS     [512]                     #

  Enable BAD BLOCK relocation?             yes                     +

  SCHEDULING POLICY for reading/writing    parallel                +

    logical partition copies

  Enable WRITE VERIFY?                     no                      +

Some of the terms used are described below:

Mirrors

On the disk placement

Mirror Scheduling Policy

Mirror Write Consistency

If you are terribly and extremely unlucky ... a crash could happen leaving the two mirrored copies of a block different.
Reading the block could give you the old or new block.
With a RDBMS, it does not mind disk is newer but must be consist when read.
MWC reminds LVM to "resilver" these recently written blocks in recovery - to make them the same.

Physical Partition (PP) Allocation

Logical Volumes striping

• From AIX 4.3.3 (not backward compatible) LVM support striping
• Often called fine striping (i.e. not PP level striping)
• Recommend 32K, 64K or 128K stripe width = 4 to 16 times I/O size most databases do 4K or 8K I/O
• Excellent for balancing serial and random I/O across disks
• Excellent read ahead serial I/O boost

Using LVM commands (with smit)

• Find unused disks: lspv
• Add them to a VG or create a VG: smitty lvm + VG + Add VG
• Check unused space in VG: lsvg <vgname>
• Create a LV: smitty lvm + LV + Add LV
• Create a Filesystem (JFS)smitty jfs + Add jfs + standard + which VG (see below)

If the command changes the VG, PV or LV then the VG is locked i.e. one command at a time.
This is needed to allow recovery of the change - if the system fails during the update.

Creating Stripes Tips:

• Stripes LV must contain a whole number of items, that is the number of PPs must divide evenly into the number of disks
• Or put another way: each disk have the same number of PPs
• Hint: use multiples of No. of disks to make this easier

LVM Create Mirrors

There are three ways to create a mirror:

1. One Step = create LV with mirror
   • These can be slow as you can have to wait until mirror complete before creating next lv
2. Create LV and later add mirror synchronise=yes
   • again slow as you can have to wait for silvering
     #Create LV and later add mirror synchronise=no and then sychronise all mirrors together
   • Use this when creating lots of lv because all mirrors can be created at the same time

So (with smit)

1. create lv's
2. add mirrors sync=no
3. syncvg -l <lvname>
   to remove "stale" partitions = resilver
   Warning:

• resilvering hurts performance
• often forgotten when recommending mirror breaking for backup purposes

LVM Higher Level Management Commands

Disown disks ready to connect them to another system

• varyoffvg <vgname>
• exportvg <vgname>
  or smitty lvm

Activate newly attached disks

• cfgmgr – finds disks
• importvg hdiskN – any one of the disks
• varyonvg <vgname> – mounts file systems

Moving LVs

migratelv (not striped)

• Use smit lvm
• Can migrate LV on a disk to another disk
• Can empty whole disk to remove it later or replac a fault disk
  Disks can be added and removed from a VG
• extendvg and reducevg
• Use smit lvm recommended

LVM rootvg good practice

Small rootvg - 1 or 2 disks (for mksysb backup)
Mirror rootvg for AIX disk protection

• Add disk hdisk1
• chvg -Qn rootvg
• For hd1 to hd8 and hd9var copy the LV
• mklvcopy hd{1,2,3,4,5,6,7,8,9var} 2 hdisk1
• syncvg -v rootvg
• bosboot -a
• bootlist -m normal hdisk0 hdisk1
  Or
• mirrorvg -s rootvg
• this second option is recommended

TIP: This command will show you formatted list of active Volume groups

for i in $(lsvg -o);do lsvg $i;done \

| awk ' BEGIN { printf("%-10s\t%10s\t%10s\t%10s\t%10s\n","VG","Total(MB)","Free","USED","Disks") };/VOLUME GROUP:/ \

{ printf("%-10s\t", $3) };/TOTAL PP/ { B=index($0,"(") + 1;E=index($0," megaby");D=E-B;printf("%10s\t", substr($0,B,D) );};/FREE PP/ \

{ B=index($0,"(") + 1;E=index($0," megaby");D=E-B;printf("%10s\t", substr($0,B,D) );};/USED PP/ \

{ B=index($0,"(") + 1;E=index($0," megaby");D=E-B;printf("%10s\t", substr($0,B,D) );};/ACTIVE PV/ { printf("%10s\t\n", $3) } '

 

Put other disks into other volume groups

• use min PP size ( 8MB or 16 MB)
• use 16 to 32 disks per volume group
• use different vg for data that might be moved elsewhere
  Avoid hot disks
• Spread data across disk with LVs or RAID5
• Mirrors add protection and performance
• Use striping whenever possible
• Think spindles not disk size
  reorgvg
• this will move physical partitions (PP) around in an attempt to get PPs in the right place
  migratepv
• this can be used to spread PPs across move disks
• also to remove all PPs from a suspect disk before replacement (due to errors being reported)

AVOIDING Disk Crashes

1. Rule One: don't let this stop your system

• RAID5 or mirror everything

1. Rule Two: monitor error logs

• Make sure you know when a disk failed

1. Rule Three: call hardware support

• That is what they are for

1. Rule Four: Don't meddle

• Only try, if you know what you are doing

1. Rule Five: Read and practise

• Get the Redbooks and try it safely

 

이 글은 스프링노트에서 작성되었습니다.