White Paper

4
4. Overview
Current SG-NFS over VxFS Support
In a Serviceguard environment, a Serviceguard NFS over VxFS (failover) package is
defined as a collection of resources including relocatable IP addresses and logical
volume groups (data disks) associated with NFS services. If the server on which a
package is running fails, then Serviceguard will automatically start the NFS package
on the adoptive node. The adoptive node takes control of the IP addresses and
disks and starts the NFS services.
CFS vs. Non-CFS Implementation
The Serviceguard A.11.18 release supports CFS which provides data and lock
coherency and allows files and filesystems to be shared and accessed concurrently
by multiple servers.
Figure 1 shows how files and filesystems are accessed differently in a non-CFS
environment versus a CFS environment. In a non-CFS environment, the highly
available filesystems move from one node to another when there is a failover. The
solid lines show which primary node provides access to which filesystem and the
dotted lines show which adoptive node provides access in the event of a failover.
The volumes are exclusively activated only on the server that currently runs the
package, which prevents files and filesystems from being altered concurrently by
multiple nodes. The Serviceguard NFS package control scripts ensure that upon
package failure or shutdown the storage is made inaccessible on the node where
the package failed or was halted.
/CFS
Storage
/mnt1
VxFS
/mnt2
VxFS
Cluster
Node A
Cluster
Node B
Figure 1. CFS Vs. Non-CFS (VxFS) Implementation
A cluster file system is concurrently accessed by all cluster nodes.
A non-CFS file system (VxFS) is exclusively activated by one node and can
transition between nodes.
In a Serviceguard CFS environment, files and filesystems are concurrently accessible
on multiple nodes. When a package fails over, the adoptive systems do not have to
mount the disks from the failed system because they are already mounted. There is a
new multi-node package that runs on each server in the cluster and exports all the
cluster filesystems. The exported filesystems do not have to be re-exported when a
package fails over. These factors may reduce failover time.