Performance analysis of the HP Serviceguard Storage Management Suite for Oracle Database

Software
For the database servers we used the following software packages:
HP-UX B.11.23.0512.034 Feature Enablement Patches for HP-UX 11i v2, December 2005
T2777BA HP Serviceguard Cluster File System for RAC plus any required
patch listed in the release notes
Oracle 9i R2 RAC 9.2.0.6 Database Server product
Software used on the client side is irrelevant for this paper.
Disk configuration
From the EVA4000 we presented 4 LUNs to the SAN network. Each LUN was 220GB size and all
protected under RAID 0+1. Cache policies selected were “mirrored” for read and “write-back” for
write.
• For Raw device based database, logical volumes were striped across the 4 LUNs and the stripe
size used was 64KB. Obviously with our selected hardware SAN topology, there are several paths
leading to the same LUN. We made sure that we selected disk device files so that I/Os were
spread across the ports to avoid any bottleneck: on the EVA side with 2 fibres for each controller
and on the server side with 2 HCA.
• For CFS, DMP takes care of the multi-pathing and recognizes the EVA4000 as an active-active
device.
The disk group was created as a shared write cluster group with no particular organization since the
EVA is taking care of the redundancy through RAID.
All the volumes were striped (stripe size 64KB) across the 4 LUNs. The filesystem was created with an
8K block size.
Database configuration
We installed the Oracle database server product on a CFS cluster mounted directory (note that it
could have been done locally too).
The 2 cluster interconnect gigabit links are also used by Serviceguard for heartbeat and CFS Data
Link Provider Interface (DLPI) traffic. The Client to Server traffic is carried over another dedicated
gigabit link.
5