LVM Link and Node Failure Recovery Time (June 2009)

Abstract
The HP-UX Shared Logical Volume Manager (SLVM) is a cluster volume manager that can support
Oracle Real Application Cluster (RAC) applications in HP-UX Serviceguard clusters.
The Using Serviceguard Extension for RAC
manual located on docs.hp.com refers to the SLVM
Recovery Time formulas provided in this document. In Using Serviceguard Extension for RAC,
the
SLVM Recovery Time is called the SLVM timeout.
Using Serviceguard Extension for RAC
requires the SLVM timeout value to calculate the Oracle
Clusterware Cluster Synchronization Services (CSS) Misscount timeout. You can optionally configure
the CSS Misscount timeout to a value lower than the default value of 600 seconds. The minimum
value is specified by the formula in Using Serviceguard Extension for RAC
, which depends on the
SLVM Recovery Time formulas provided in this document.
You can also use the SLVM Recovery Time formulas to estimate SLVM node failure recovery times in
HP-UX Serviceguard cluster RAC application environments.
This document also provides the LVM Link Failure Recovery Time formula for estimating LVM Physical
Volume link (PV Link) failure recovery times. The formula applies to either standalone (non-cluster) or
Serviceguard cluster environments.
In addition, this document provides general formulas for estimating SLVM node failure recovery times
and LVM PV link failure recovery times. This document is not intended to serve as a comprehensive
guide for configuring SLVM or LVM to optimize these recovery times.
Formulas
SLVM Recovery Time Formula
The SLVM Recovery Time is the time that SLVM takes to recover from a node failure in a SGeRAC
cluster. The SLVM Recovery Time formula is computed based on the following factors, depending on
the version of HP-UX used:
• The maximum physical volume timeout (PV timeout).
• The number of mirrored logical volumes (LVs).
• The number of shared volume groups (VGs) with mirrored LVs.
The formula is in seconds.
The formula applies even if there are no mirrored LVs in any of the volume groups in the cluster.
A shared VG is a volume group that was activated in shared mode in a SGeRAC cluster using the
vgchange –a s command.
The formula uses the following variables:
Max PV Timeout
The maximum physical volume timeout in seconds for all PVs in all of the shared VGs in the cluster.
The smallest Max PV Timeout setting that should be used in the formulas is the default PV
timeout value of 30 seconds. The PV timeout setting for a PV is displayed by the pvdisplay
command.
number of mirrored LVs
The number of mirrored logical volumes in all shared VGs in the cluster that have consistency
recovery set to either MWC (recovery through Mirror Write Cache) or NOMWC (recovery without
Mirror Write Cache). You can display the consistency recovery setting for an LV with the
lvdisplay command. Mirrored logical volumes that have consistency recovery set to NONE (no
recovery) do not factor into the formulas.
number of shared VGs with mirrored LVs
2