Cloud DR and Backup for a 30 TB Data Warehouse
A retail chain runs a 30 TB Oracle data warehouse on a Huawei OceanStor array at its headquarters. The company wants disaster recovery in a Huawei Cloud region and 30 daily backup copies. The available IP WAN link is 100 Mbps, shared with business traffic. The peak change rate is 2% of the data per hour, and the business requires RPO 15 minutes and RTO 4 hours. The initial copy over the WAN is estimated at 28 days. The CIO asks whether HyperReplication synchronous, HyperReplication asynchronous, or backup to OBS is the right choice, and how to make the design achievable without upgrading the link. A temporary 1 Gbps link can be rented for two weeks at extra cost.
1. Assess whether the 100 Mbps link can meet the 15-minute RPO with replication. Calculate the required bandwidth and recommend a replication mode.
Read the worked answer
Calculate the change volume: 2% of 30 TB = 600 GB per hour. For a 15-minute RPO, 150 GB must cross the link in 900 s. Required throughput = 150 GB x 8 / 900 s = 1.33 Gbps, before overhead and compression. The 100 Mbps link delivers about 0.1 Gbps, or 45 GB/h, so it cannot keep up with 600 GB/h even continuously; the backlog grows and the 15-minute RPO is impossible. Synchronous replication is also unsuitable over a WAN because it adds round-trip latency to every write.
Recommendation: use HyperReplication asynchronous for DR, but first seed the initial 30 TB over the temporary 1 Gbps link, about three days at line rate and less with compression, or use physical media. Keep the link for incremental replication, or upgrade to at least 1.5 to 2 Gbps and enable dedupe/compression. If the link cannot change, renegotiate the RPO: at 100 Mbps the sustained change rate exceeds the link, so even a multi-hour RPO is not sustainable without reducing the change volume.
2. Design the backup copy scheme for 30 daily copies using cloud services, including retention and cost controls.
Read the worked answer
Use a 3-2-1 scheme: production array, DR replica in the cloud region, and 30 daily backups in OBS. Configure eBackup or OceanProtect to take a weekly full and daily incremental; use source-side dedupe and compression and encryption in transit and at rest. Set an OBS lifecycle rule to move backups older than 7 days to an archive class, but keep 30 daily restore points. Enable WORM or immutable retention for at least the latest 7 copies to resist ransomware. Throttle backup traffic during business hours and run the full backup over the temporary high-bandwidth link or during the initial seed. Test a restore of a sample database and a full VM to prove the backup is usable, not just completed.
3. What risks and acceptance tests should be included in the DR runbook?
Read the worked answer
DR runbook tests should cover: measured RPO under peak change; failover of the database and dependent applications; DNS and connection-string changes; RTO of 4 hours; failback; split-brain prevention; encryption-key availability in the cloud; and monitoring and alerting. Include a decision tree for declaring disaster, a contact list, and a rollback to the HQ array if the cloud region is unavailable. Acceptance requires an actual failover exercise with business sign-off, not a paper test.