Issue #9 July 2005

Red Hat Speaks

Heinz Mauelshagen, Consulting Development Engineer, Cluster and Storage Development

This month's issue of Red Hat Magazine features Red Hat storage solutions: Red Hat Global File System, LVM2, and Red Hat Cluster Suite. We interviewed Heinz Mauelshagen, one of Red Hat's LVM developers, to gain more in-depth knowledge of the Linux Logical Volume Manager as well as learn about the future of LVM.

For more information on LVM, refer to this month's article The Linux Logical Volume Manager.

What is the most important advantage in using LVM over raw partitions when I configure my Red Hat® Enterprise Linux® server?
The most important advantage is LVM's ability to resize on the fly. A system administrator can resize volume groups and logical volumes on the fly without shutting down applications or rebooting the system, hence completely avoiding costly downtime.
Can I improve the performance of my storage solution using LVM?
Yes. LVM provides the ability to create striped logical volumes. This striping enables parallel access to multiple disks, adding up their respective throughput and improving I/O performance.
Can I improve the availability of my Red Hat Enterprise Linux servers with LVM?
Yes. We are implementing software mirroring (i.e. RAID 1) at the Logical Volume level. Existing logical volumes can be changed to mirrored ones while the disks are online, avoiding downtime.
Why should I use LVM over the Linux MD driver? Can I use them together?
LVM satisfies the requirement for fine grained storage control and online data relocation wheres MD handles grouped devices of complete disks or disk partitions without the ability to relocate or resize online. LVM1 and LVM2 can be used either on top of or below MD devices.
I heard there is a new version of LVM, is that true? If so, what are some of its advantages?
Yes. We have created a version 2 of LVM (LVM2) with enhanced resilience (such as new transaction metadata format) and user configurability (such as device name filters, metadata backup and archive paths, metadata archive history, etc.), new output display tools, and many performance optimizations (such as asynchronous snapshots). Refer to this month's article on LVM2 for more details.
Additions include the upcoming aforementioned software mirroring support.
LVM2 fully supports the on-disk LVM1 metadata format (read and write) as well as conversion of the older format to the new resilience-enhanced LVM2 format via the new vgconvert tool.
How scalable is LVM? Can I manage hundreds of disks with it?
Yes. LVM1 scales to hundreds of disks and LVM2 supports thousands. Of course, a practically unlimited amount of volume groups and logical volumes can be created on the 2.6 kernel with LVM 2 as well (appropriately 17K logical volumes have been tested).
How does LVM compare to other Linux volume managers such as Veritas VxVM?
LVM comes with many but not all of the features included in the more expensive, commercial volume managers. The main shortfall LVM still has is support for higher RAID levels (RAID 1,3,5,6), combinations of those, and remote replication capabilities.
As mentioned, RAID1 support is already in the works.
Can I migrate data from a failing physical disk to another disk using LVM? Can I continue to use that disk while I am performing this migration?
Yes. The pvmove command assists such online data relocation, which is mandatory to getting as much live data off a failing drive as possible (failure management). Once we have released RAID1 support, this will of course be recommended to be resilient against failures of single disks.
Another administrable benefit of pvmove is to relocate data off an old drive so it can be removed from the system afterwards (performance/configuration management). During the life migration, full access to the data is granted.
How do snapshots work in LVM? How many snapshots can I have on a single LVM volume before performance starts to degrade?
Snapshots are instant 'frozen' live copies (think just another logical volume which gains access to a copy) of any given logical volume by housekeeping any changes that happen to either of them. Read/write access to both is still possible (read-only in LVM1).
In theory, any number of snapshots can be created for a particular logical volume. For example, periodical snapshots might be useful for creating consistent backups or run tests on live data avoiding damage to the original data.
In practice, every snapshot taken degrades the access performance to the origin because of the necessary housekeeping (we're working on performance optimizations).
Depending on the application, just a few snapshots might saturate throughput in heavy loaded cases. With lighter loads, more than that is possible.
What other new features are coming out in LVM this year?
Upcoming LVM features include:
  • The before-mentioned RAID1 support for logical volumes, including the ability to 'upgrade' exiting LVs to be mirrored and to do split mirrors.
  • Ability to have the root directory on a Logical Volume, including a mirrored root directory