Quick Take
Quick Take: ZDLRA RA26 protects Exadata and Oracle Database estates with real-time redo and incremental-forever virtual fulls. CapEx starts at 2 compute + 3 storage, then grows storage cells (RA26 or RA26-Z) to 17 per rack. Freeze usable TB, redundancy, network, and PDU on one BOM before order.
Base Rack (= 2 Compute + 3 Storage)
Quick Answer: Every RA26 rack keeps two compute servers for Recovery Appliance software HA. Capacity starts at three storage servers and expands by adding storage servers only — not by adding compute nodes. Plan client network and PDU with the base rack, not as afterthoughts.
Base infrastructure (every RA26 / RA26-Z rack):
- 2× Recovery Appliance compute servers (fixed count)
- 3× storage servers minimum (RA26 or RA26-Z class)
- 2× RoCE Network Fabric Switches
- 2× redundant PDUs (phase/voltage matched to site)
- 1× management network switch
- Client access network for protected databases (10/25 GbE typical; 100 GbE options)
Optional: Fibre Channel HBAs for tape/SAN adjacency where compliance requires it.
When This Checklist Matters
Quick Answer: Use this page when you already chose ZDLRA for Oracle protection and need every CapEx line frozen — especially beside an Exadata refresh. Skip it if you still need ZDLRA versus RMAN versus third-party platform selection, or only need RA26 hardware specs depth.
- Platform bake-off: ZDLRA vs RMAN vs Third-Party — decide if ZDLRA is the right path.
- Specs depth: ZDLRA RA26 Specs — architecture and ordering reference.
- Production rack BOM: Exadata X11M Elastic BOM Checklist — size the estate ZDLRA protects.
- Exadata refresh: X9M/X10M to X11M Upgrade — common trigger to add or resize ZDLRA.
RA26 vs RA26-Z — Usable Capacity
Quick Answer: Oracle publishes RA26 and RA26-Z storage-server classes with different usable TB steps. Under normal redundancy, a base RA26 rack is about 332 TB usable; each added RA26 cell about 112 TB. RA26-Z base is about 164 TB; each added cell about 56 TB. High redundancy roughly halves those figures — always state mode on the RFQ.
| Published guidance (normal redundancy) | RA26 | RA26-Z |
|---|---|---|
| Base rack (3 storage) | ~332 TB usable | ~164 TB usable |
| Per added storage server | ~112 TB usable | ~56 TB usable |
| Full rack (17 storage) | ~1.9 PB usable | ~943 TB usable |
| High redundancy (approx.) | ~72 TB / cell; ~1.2 PB full | ~36 TB / cell; ~605 TB full |
| Multi-rack cluster | Up to 14 full racks (~26.6 PB) | Up to 14 full racks (~13.2 PB) |
Sizing rule of thumb
Model protected database size × retention × change rate, then apply incremental-forever / virtual-full efficiency — do not size ZDLRA as if every full backup is stored every night. Confirm policy retention and replication to a second ZDLRA before freezing cell count.
Network, PDU & Scale-Out
Quick Answer: Protected databases reach ZDLRA on a dedicated client Ethernet path so redo and backup traffic stay off production LAN contention. Internal RoCE stays inside the appliance. Wrong PDU phase or undersized client bandwidth is a common delivery delay after CapEx approval.
| BOM decision | What to freeze |
|---|---|
| Client network | 10/25 GbE SFP28 baseline; dual-port 100 GbE when estate throughput requires it |
| Internal fabric | Dual RoCE switches (factory) — do not omit from multi-vendor quotes |
| FC option | 32 Gb FC HBAs only if tape/SAN adjacency is on the brief |
| PDU | 1-phase or 3-phase; HV/LV; region plug standard |
| In-rack growth | Add storage servers toward 17; follow Oracle population rules |
| Multi-rack | Cluster up to 14 full racks when single-rack usable TB is insufficient |
BOM Line Items to Freeze
Quick Answer: Treat ZDLRA CapEx as seven hardware and service buckets plus a separate protection-policy brief for databases. Keep the Exadata production BOM on its own CapEx column — recovery usable TB is not a substitute for Smart Scan cells or database-server memory.
- Model family: ZDLRA RA26 base / factory rack integration
- Compute: 2× compute servers (standard on every rack)
- Storage: count × RA26 or RA26-Z; state normal vs high redundancy
- Fabric: 2× RoCE switches + management switch
- Client / optional FC: adapters, optics, cabling to Exadata/client network
- Power: 2× PDUs — phase, voltage, region
- Services: Premier Support for Systems, configuration/ACS, shipping
- Adjacency (separate CapEx): Exadata elastic BOM; optional second ZDLRA for replication
| Example shapes | Storage servers | Notes |
|---|---|---|
| Entry / base | 3 | Minimum; pick RA26 or RA26-Z + redundancy |
| Mid growth | 6–9 | Common step after first retention year |
| Near-full rack | 12–17 | Confirm power/cooling and population slots |
| Multi-rack | Cluster when single-rack usable TB or throughput ceiling binds | |
RFQ Checklist — Beside Exadata CapEx
Quick Answer: One data protection brief should equalize estate size, recovery objectives, retention days, RA26 class and redundancy mode, storage server count with a growth path, client network class, power distribution and lead time. Put Exadata production and ZDLRA recovery as two capital columns sharing identical protected-database license assumptions.
- Estate: database count/size, Exadata vs standalone, change rate, retention days
- RPO/RTO: real-time redo requirement vs backup-cycle only
- Class: RA26 vs RA26-Z; normal vs high redundancy
- Counts: storage servers now + 3–5 year growth path
- Network: client GbE class; FC yes/no; optics/cabling
- Power: PDU phase/voltage; rack RU and cooling
- Replication: second ZDLRA / remote site yes/no
- Exadata column: link to elastic BOM if production refresh is concurrent
- Delivery: factory lead time, install, protection-policy cutover window
Frequently Asked Questions
Quick Answer: Start from the 2+3 base, pick RA26 or RA26-Z with an explicit redundancy mode, then freeze client network and PDU. Size usable TB from retention and change rate — not from nightly full-backup folklore — and keep Exadata production CapEx on a separate line.
Two compute servers and three storage servers, plus dual RoCE switches, redundant PDUs, and a management switch. Grow capacity by adding storage servers up to 17 per rack.
RA26 for larger usable TB per cell; RA26-Z for smaller capacity steps. Always state normal vs high redundancy — high redundancy roughly halves usable capacity.
No. ZDLRA is the recovery plane. Exadata cells serve production Smart Scan I/O. Budget both racks when refreshing a mission-critical Oracle estate.
Oracle documents that replacing older storage servers with newer ones while keeping existing backups online is not supported without re-image. Plan generation swaps as migration projects, not hot swaps.
Submit one RFQ with the BOM checklist above.
Key Takeaway
Key Takeaway: ZDLRA RA26 CapEx wins or loses on usable-TB honesty — RA26 vs RA26-Z, redundancy mode, storage count, and network/PDU completeness — beside a separate Exadata production BOM. BENQUAN GLOBAL LIMITED supplies Oracle ZDLRA RA26 with competitive pricing and configuration support for global buyers.