PowerProtect DD Architecture and Hardware Components
What this domain covers
The architecture domain tests whether you can explain how a PowerProtect DD appliance converts a backup stream into deduplicated, protected data, and whether you can map a customer requirement onto the right controller, shelf and port configuration.
The inline data path
Every front end (DD Boost, NFS/CIFS, VTL, NDMP) feeds the same pipeline:
- Segmentation - the stream is split into variable-length segments, typically averaging around 8 KB, so small shifts in the data do not invalidate every downstream segment.
- Fingerprint and lookup - each segment is fingerprinted and checked against the global index. Duplicates become pointers; only unique segments continue.
- Local compression - unique segments are compressed (default
lz;gzvariants trade CPU for a better ratio). - Optional encryption - AES-256 is applied after compression, so encrypted data still deduplicates and compresses.
- Landing - data is acknowledged into NVRAM, then committed to RAID-protected disk inside a log-structured file system.
Data Invulnerability Architecture
DIA is a set of mechanisms, not one feature:
- End-to-end checksums on data and metadata plus write verification on every write to disk.
- Copy-on-write snapshots that give the file system a consistent restart point, so an unclean shutdown resumes rather than rebuilds.
- Continuous fault detection and healing (background scrubbing) that finds and repairs bad blocks from RAID parity before a restore ever reads them.
- Fast restore, which reconstructs only the stripes actually needed instead of a whole RAID group.
- Capacitor/battery-backed NVRAM so acknowledged writes survive power loss.
Hardware families
- DD3300 - entry-level 2U all-in-one for remote or branch offices; no external shelf expansion.
- DD6400 / DD6900 / DD9400 / DD9900 - current scale-up families (smallest to largest controller), with SAS-attached external shelves for capacity growth and multiple 10/25 GbE data ports.
- All controllers carry NVRAM, RAID controllers, hot spares and redundant power and fans. A fully populated shelf weighs far more than an empty chassis, which affects rail and lift planning.
Worked example
A nightly full of 500 GB is kept 30 days, so 15 TB logical. Global dedupe against previous fulls plus compression leaves 1.2 TB physical. Reduction is 15 TB / 1.2 TB = 12.5:1. A planner who mistakenly sized on the 1.2 TB physical figure would buy a system that cannot hold 30 days of data; always size from the post-reduction number and then add growth and free-space headroom.
Traps that cost marks
- Calling DD Boost server-side dedupe - it is client-side distributed segment processing.
- Multiplying dedupe and compression factors again when the system already reports a combined reduction ratio.
- Confusing logical (pre-dedupe) capacity with usable capacity.
- Assuming every model accepts external shelves.