Access Official Technical PDF

Share Summary:LinkedInEmail

For organizations running large Oracle databases, backup design affects more than data protection: it can influence production workload, recovery time and the complexity of infrastructure changes. Dialog Semiconductor’s ZDLRA backup architecture paired an Oracle Zero Data Loss Recovery Appliance with an Exadata standby system, using real-time redo shipping to move backup activity away from the primary database. The deployment also supported recovery and migration of a production database of approximately 52 TB.

ZDLRA backup architecture: shifting work off the primary

Dialog’s earlier environment used an Exadata X4 primary cluster and an Exadata X2 standby cluster, alongside application servers running on VMware ESX and NFS file systems. Backups relied on an NFS-mounted backup device. In the later setup, an Exadata Cloud at Customer (ExaCC) X6 system served as the primary, with the Exadata X4 retained as the standby and ZDLRA added as the backup appliance.

The key architectural choice was to avoid sending backup traffic directly from the primary. Online redo was shipped from the ExaCC primary to the standby using redo routes. The standby then sent redo to the on-premises ZDLRA. In this arrangement, the primary database did not carry the backup load, while the appliance received the information needed to support recoverable backups.

The broader environment also included Oracle’s control plane and virtual machines, with a longer-term goal of moving services into Oracle’s public cloud at the customer’s site. The design therefore connected existing on-premises systems with ExaCC rather than treating backup as an isolated appliance deployment.

Recovery improvements and backup operations

The reported recovery metrics show how the change affected both recovery time and backup duration. Before ZDLRA, the recovery time objective (RTO) was 48 hours. By April 13, after the appliance was in use, the reported RTO had fallen to 12 hours—about four times faster.

Backup processes also changed. A weekly Level 0 (L0) backup had previously taken 38 hours. The first L0 with ZDLRA took about 10 hours. After that, the primary no longer needed to produce L0 backups: ZDLRA could create its own virtual L0 backups. Incremental Level 1 (L1) backup and archive log backup had previously taken more than three hours; the reported incremental L1 duration with ZDLRA was around 1.5 hours, with real-time redo shipping enabled.

These figures describe the deployment’s reported results, not a universal performance guarantee. They illustrate why backup topology matters: routing redo through a standby can reduce backup activity on a production system, while appliance-managed virtual backups can change the work required from the database environment.

Using ZDLRA and Data Guard for database migration

The same components were used to support a move to a new environment. The process began by backing up the database to ZDLRA while enabling redo log shipping. A virtual full backup could then be restored to the target environment at a selected point in time. The restored database was added as a standby, and Oracle Data Guard was used to switch over to it.

For a production database of approximately 52 TB, the team used RMAN duplication from ExaCC database nodes to restore from ZDLRA. The Data Guard switchover to the restored database took less than five minutes. The approach allowed the database to be prepared on the new environment before the switch, without placing the migration’s backup load on the primary.

For infrastructure teams planning a similar process, the important point is the sequence: establish backup and redo movement, restore the target database, make it a standby, and then perform the Data Guard switchover. Each step depends on validating restore and recovery behavior in the target configuration.

Early deployment challenges and validation

Initial RMAN duplication and restore testing exposed two timeout-related problems. One involved a network switch port failure, which Oracle Support helped identify and the team corrected. Another occurred when restoring a bigfile datafile of about 12 TB: the client-side operation timed out because of an issue with the restore servlet. The issue was addressed by changing an internal ZDLRA parameter and was fixed in a later software release.

ZDLRA development was involved in follow-up, and findings from the early deployment were incorporated into later software. The team repeatedly tested manual restore and recovery as well as RMAN duplication, recovering the 52 TB production database with different RMAN attributes and channel tuning. The purpose was to confirm restore speed while checking that duplication and recovery completed without failures.

This experience highlights a practical implementation requirement: large-scale recovery plans should be tested under realistic conditions, including the database size, network path, RMAN configuration and restore method. Successful backup completion alone does not establish that a full recovery or migration will meet operational needs.

Operational implications for recovery planning

The deployment’s reported benefits included a lower RTO, an RPO below one second with real-time redo shipping, and less backup work on the primary. The architecture used forever incremental Level 1 backups, with a logical incremental Level 0 available for restore. Dialog also reported that ZDLRA did not require special monitoring or maintenance from its team because it was an Oracle-managed appliance. Appliance status and activity could be reviewed through business intelligence reports and existing Enterprise Manager 13 monitoring.

For plant and IT operations leaders, the wider lesson is to evaluate backup architecture as part of service continuity and infrastructure planning. A standby system can provide a path for redo to reach the recovery appliance without sending backup load to the primary, while ZDLRA and Data Guard can support both recovery and carefully staged migration. The reported reductions in recovery and backup duration show the potential value of this design; thorough restore testing remains essential to confirm that it works for a specific environment.

Access Official Technical PDF

About The Author