•Cluster arbitration: dual cluster lock disks (Logical Volume Manager (LVM) volume groups) or Quorum Server.

–Note: Three-site configuration is supported using Quorum Server to prevent split-brain, the condition in which two equal-sized subgroups of cluster nodes re-form a cluster independent of each other.

•Nortel OPTera DWDM switches were used for initial testing and certification.

Supported software configuration

•Operating system: HP-UX 11.11 or HP-UX 11i v2.0

•Shared Logical Volume Manager (with physical volume links and HP MirrorDisk/UX, HP StorageWorks SecurePath with EVA storage)

•VERITAS VxVM/CVM 3.5 (with DMP and mirroring) for cluster volume management

•HP Serviceguard Quorum Server or dual cluster lock disks for cluster arbitration (The dual cluster lock disks are required—one at each center—to facilitate recovery from an entire data center failure.)

•HP Serviceguard and HP SGeRAC A.11.15 or later

•Oracle RAC 9.2 or later

•HP StorageWorks Extended Fabric on Fibre Channel switches (HP StorageWorks Extended Fabric enables dynamically allocated long-distance configurations in a Fibre Channel switch.)

Test descriptions

The following categories of testing were performed:

•IPC (Inter-Process Communication) tests

•I/O tests

•Online transaction processing (OLTP)-like workload tests

•Failover tests

IPC test description

Raw IPC throughput was evaluated using CRTEST, a micro-level performance benchmark. The test first updates a set of blocks in a hash cluster table on instance A, and then an increasing number of clients running SELECT are started on instance B. These queries cause messages to be sent from instance B to instance A; instance A returns a CR block. CR Fairness Down Converts were disabled for this test to create a fundamental dialog of “send a message” and “receive a CR block” back.

The CRTEST (IPC) tests were performed initially using one Gigabit Ethernet network for cluster interconnection, and then the tests were repeated using two interconnects.

I/O test description

The I/O tests were executed using the Diskbench (db) disk subsystem performance measurement tool. The db tool measures the performance of a disk subsystem, host bus adapter, and driver in terms of throughput (for sequential operation) and number of I/Os (for random operation). Diskbench can evaluate the performance of kernel drivers and can be used on one-way or multiprocessor systems to completely saturate the processors and effectively measure the efficiency of a disk subsystem and associated drivers. The tests were performed using a mix of 60% reads and 40% writes to simulate “real-world” traffic.

OLTP test description

The industry-standard TPC-C test was utilized to simulate OLTP transactions over the cluster interconnects and DWDM network. TPC-C can emulate multiple transaction types and complex database structures. The benchmark involves a mix of five concurrent transactions of varying type and

10