Serviceguard Extended Distance Cluster (EDC) with VxVM/CVM Mirroring on HP-UX, May 2008

This command changes the size of vol1 to 20 GB. At the same time, it will also change the file

system size to 20GB, if vol1 contains a file system. The associated DCL volume will also be resized

to accommodate the larger FMR. The vxresize command can also be used to shrink a volume. For

more details on resizing VxVM/CVM volumes, please see the VERITAS Volume Manager

Administrator’s Guides referred to in the Related Documents

section.

In some circumstances, VxVM converts a volume from a non-layered volume into a layered volume

when resizing it. Use vxassist -g dgEDC convert vol1 mirror=enclr to convert it back to

a non-layered volume.

To guarantee that the volumes remain highly available (with complete mirrors at both

sites/enclosures), it is a good practice to always specify storage attributes for resizing, re-layout, and

conversion operations on VxVM volumes and verify the locality of the storage objects afterwards.

When creating new volumes in existing disk groups, the same recommendations apply as described

above and as shown in the example in figure 7.

VxVM/CVM 5.0 Tip:

The “site-awareness” feature allows configuring “site consistency” for disk

groups. If a volume is created in a disk group which is site consistent the

volume inherits this setting. Together with the mirror directive

mirror:site site:DC1 site:DC2 VxVM will ensure that one plex is

located at each site when a new volume is created.

It is a good practice, to verify the location of the individual plexes after volumes are created or

extended.

EDC specific failure scenarios

An EDC protects against additional failures, as compared to a local Serviceguard cluster, due to the

EDC architecture. The following scenarios are typical failure cases that a local Serviceguard cluster

will not protect against, but an EDC will:

Complete site failure

Protecting against a complete site failure is the greatest value an EDC offers over a local Serviceguard

cluster. A complete site failure removes half of the cluster components (nodes and storage systems)

and leaves the other half, residing in a different data center, available.

Site failure simulation and reaction

There are several ways to simulate a complete site failure:

• Perform a power failure of the entire site. This is probably the most realistic (and drastic) simulation

since it removes all cluster components from that site. This includes the nodes, storage arrays,

network switches, and FC switches.

• Perform a hard failure of all nodes at a site with a reset or transfer of control (TOC), and disable all

FC switches at the same site. This will remove the nodes and storage arrays from the cluster.

• Perform a failure of all nodes at a site by rebooting them and simulating a storage failure by

disabling the FC switch ports on the failing site. This removes access to the storage array at the

failing site from the nodes still running at the remaining site.

The information in this white paper is based on site failure tests that have been simulated by TOCing

all nodes at one site and then disabling the FC switch on that site.