Developing a Volume Group Strategy

Disk failure is the most common hardware failure in the storage system, followed by failure of adapters and power supplies. Protection against disk failure primarily involves the configuration of the logical volumes (see "Developing a Logical Volume Strategy" ). However, volume group size also plays a part as is explained below.

To protect against adapter and power supply failure, you need to consider a special hardware configuration for any specific volume group. Such a configuration includes two adapters and at least one disk per adapter, with mirroring across adapters, and a nonquorum volume group configuration. The additional expense of this configuration is not appropriate for all sites or systems. It is recommended only where high (up-to-the-last-second) availability is a priority. Depending on the configuration, high availability can cover hardware failures that occur between the most recent backup and the current data entry. High availability does not apply to files deleted by accident.

Prerequisites

It is important that you understand the material contained in the "Logical Volume Storage Overview" .

When to Create Separate Volume Groups

You may want to organize physical volumes into volume groups separate from rootvg for the following reasons:

• For safer and easier maintenance.
  • Operating system updates, reinstallations, and crash recoveries are safer because you can separate user file systems from the operating system so that user files are not jeopardized during these operations.
  • Maintenance is easier because you can update or reinstall the operating system without having to restore user data. For example, before updating, you can remove a user-defined volume group from the system by unmounting its file systems, deactivating it (using varyoffvg), then exporting the group (using exportvg). After updating the system software, you can reintroduce the user-defined volume group (using importvg), then remount its file systems.
• For different physical-partition sizes. All physical volumes within the same volume group must have the same physical partition size. To have physical volumes with different physical partition sizes, place each size in a separate volume group.
• When different quorum characteristics are required. If you have a file system for which you want to create a nonquorum volume group, maintain a separate volume group for that data; all of the other file systems should remain in volume groups operating under a quorum.
• To have multiple JFS logs or JFS logs dedicated on one physical volume for the purpose of reducing bottlenecking, especially on server machines.
• For security. For example, you might want to remove a volume group at night.
• To switch physical volumes between systems. If you create a separate volume group for each system on an adapter that is accessible from more than one system, you can switch the physical volumes between the systems that are accessible on that adapter without interrupting the normal operation of either (see the varyoffvg, exportvg, importvg, and varyonvg commands).
• To remove disks from the system while the system continues to run normally. By making a separate volume group for removable disks, provided the volume group is not rootvg, you can make removable disks unavailable and physically remove them during normal operation without affecting other volume groups.

High Availability in Case of Disk Failure

The primary methods used to protect against disk failure involve logical volume configuration settings, such as mirroring. While the volume group considerations are secondary, they have significant economic implications because they involve the number of physical volumes per volume group:

• The quorum configuration, which is the default, keeps the volume group active (varied on) as long as a quorum (51%) of the disks is present. For more information about quorum requirements, see the section on vary-on process in the "Logical Volume Storage Overview" . In most cases, you need at least three disks with mirrored copies in the volume group to protect against disk failure.
• The nonquorum configuration keeps the volume group active (varied on) as long as one VGDA is available on a disk (see "Changing a Volume Group to Nonquorum Status" ). With this configuration, you need only two disks with mirrored copies in the volume group to protect against disk failure.

When deciding on the number of disks in each volume group, you also need to plan for room to mirror the data. Keep in mind that you can only mirror and move data between disks that are in the same volume group. If the site uses large file systems, finding disk space on which to mirror could become a problem at a later time. Be aware of the implications on availability of inter-disk settings for logical volume copies and intra-disk allocation for a logical volume.

High Availability in Case of Adapter or Power Supply Failure

To protect against adapter or power supply failure, depending on the stringency of your requirements, do one or more of the following:

• Use two adapters, located in the same or different cabinets. Locating the adapters in different cabinets protects against losing both adapters if there is a power supply failure in one cabinet.

• Use two adapters, attaching at least one disk to each adapter. This protects against a failure at either adapter (or power supply if adapters are in separate cabinets) by still maintaining a quorum in the volume group, assuming cross-mirroring (copies for a logical partition cannot share the same physical volume) between the logical volumes on disk A (adapter A) and the logical volumes on disk B (adapter B). This means that you copy the logical volumes that reside on the disks attached to adapter A to the disks that reside on adapter B and also that you copy the logical volumes that reside on the disks attached to adapter B to the disks that reside on adapter A as well.

• Configure all disks from both adapters into the same volume group. This ensures that at least one logical volume copy will remain intact in case an adapter fails, or, if cabinets are separate, in case a power supply fails.

• Make the volume group a nonquorum volume group. This allows the volume group to remain active as long as one Volume Group Descriptor Area (VGDA) is accessible on any disk in the volume group (see "Changing a Volume Group to Nonquorum Status" ).

• If there are two disks in the volume group, implement cross-mirroring between the adapters. If more than one disk is available on each adapter, implement double-mirroring. In that case, you create a mirrored copy on a disk that uses the same adapter and one on a disk using a different adapter.

Decide on the Size of Physical Partitions

The physical partition size is set when the volume group is created. The default size is 4MB. The default is designed to suit most sites and systems but may not be appropriate in every case. You can choose a partition size as small as 1MB to gain flexibility in sizing but this requires more partitions. The additional partitions create more overhead for the Logical Volume Manager (LVM) and are likely to affect performance.

If you make the partitions larger than 4MB, you lose some sizing flexibility and may also waste space. For example, if you have 20MB partitions, then your JFS log will have to be 20MB when it only needs 4MB. Some waste may be an acceptable tradeoff if the particular site or system requires larger partitions.

Note that you may only create and extend physical partitions in increments that are a factor of their size; for example, 20MB partitions are created or extended in 20MB increments.