5

A co-worker asked me why we should use LVM. With my limited LVM knowledge, I said because it allow you to easily resize/manage volumes!

His idea is this:

We use ESX, and have the ability to increase the size of a disk. Instead of using LVM he proposed this scenario:

  1. increase /dev/sdb by 100GB in vsphere client
  2. back in Linux rescan /dev/sdb: echo 1 > /sys/block/sdb/device/rescan
  3. resize the FS on-line: resize2fs /dev/sdb

Done.

This seems fine. I'm not sure though. What's wrong with this scenario as opposed to using LVM's? What's a better option for allowing disk expansion?

3 Answers 3

5

LVM is great for that when you're not in a virtual environment. When the disk can be cloned and grown from outside the environment, that benefit is greatly reduced.

2

I wouldn't bother using LVM in VMWare solutions. There's some overhead, it's a bit redundant and would require a redesign of your current system. I think this goes back to planning, but you should have an idea of the disk space needs prior to building the server. At least with VMWare, it's extremely easy to add space, as you've noted.

I personally avoid using LVM. Maybe it's an old-school approach, or maybe I'm spoiled by HP SmartArray RAID controllers, but I find that I can predict and plan disk utilization needs based on the application requirements and environment. Most systems have one or two "growth partitions", depending on the application. It may be /home for a multiuser system. /opt for an application server. /var for a system that's logging-heavy or /var/lib/pgsql for a database server... The partition that will need to grow should be the last partition in the table or on its own hardware volume.

Also see: LVM dangers and caveats

There's a purpose for multiple partitions, based on the Filesystem Hierarchy Standard. In the example below, /var and /appdata have the most activity. /appdata is setup to grow, if necessary. /var has enough headroom for cyclical logging activity. /usr and /boot don't change often. / may grow a bit, but if it becomes an issue, mount another filesystem under that hierarchy. That's fairly consistent across OS distributions.

Filesystem            Size  Used Avail Use% Mounted on
/dev/cciss/c0d0p2     9.7G  3.4G  5.9G  37% /
/dev/cciss/c0d0p7     996M   34M  911M   4% /tmp
/dev/cciss/c0d0p6     3.0G 1023M  1.8G  37% /var
/dev/cciss/c0d0p3     5.9G  2.0G  3.6G  37% /usr
/dev/cciss/c0d0p1      99M   23M   71M  25% /boot
/dev/cciss/c0d0p8      79G  6.5G   73G   9% /appdata
tmpfs                 2.4G     0  2.4G   0% /dev/shm

1

Another case for LVM, even on a virtual host: it allows you to extend a filesystem across multiple disks or disk sets.

Suppose you're running ESX without a SAN, and run out of space in the datastore holding a particular VM that needs more space; if you add more physical disks, they will typically create new datastores, which you can add virtual disks to; those virtual disks can be added as new Physical Volumes to your LVM Volume Group.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .