fsck, pass number and lvm

[Fabien GUTIERREZ]

after a system crash to speed up the system check you can use the parallel feature of fsck. by specifying different pass number (/etc/fstab) for different hard drives you can use this feature. is the same possible using Logical Volumes in LVM ? Did some of you had some experiences with such parameter optimizing.

[Simon Hargrave]

The pass number is for filesystems, not "hard disks" or "logical volumes" per se. As far as fsck is concerned hard disks and logical volumes are one and the same.

Just bear in mind the logical volume's physical location, eg fscking 10 LV's on one PV at the same time will be slower than fscking 10 LV's on 10 separate PV's.

[Steven E. Protter]

Shalom Fabien,

I recommend against this. The benefit due to disk layout and i/o contention is negligible and it violates a simple principle of not changing procedures that assure system integrity.

SEP

[Bill Hassell · Accepted solution]

fsck uses the pass number in /etc/fstab to start multiple copies of itself. For HFS filesystems, fsck is very slow and very disk intensive for each filesystem. So it makes a lot of sense to run fsck in parallel on filesystems connected to different I/O cards. When fsck takes several minutes to complete for each filesystem, running several fsck's on the same channel will really slam the disk with large seeks.

Now for VxFS filesystems, fsck takes just a few seconds per filesystem so the need for separate passes is significantly reduced. On a small server or workstation, fsck will not be much of an issue. But on a server with dozens or even hundreds of mountpoints, creating separate passes does make sense, even when fsck needs only a few seconds. Even so, the fsck passes should not be more than a minute or two during a forced (not clean) reboot. And on a properly patched system that has proper power protection, an unexpected reboot may only occur every couple of years or longer.

[Devender Khatana]

Hi,

This is not much important in present scenarios where the LVOLs having VxFS file systems only take a few seconds each to complete this process. We have systems having upto 60 mount points and in case of one such situation it only takes a max of 2-3 minutes in doing the whole mounting causing us not to think about tuning such parameters.

HTH,
Devender

[Fabien GUTIERREZ]

Seems like this tuning won t improve much on vxfs the disk check.