Skip to content

ADR-053: Application-level DRR degradation as CSV subcategory variants

Date: 2026-02-20 Status: Accepted

Context

PowerStore's published DRR benchmarks assume reducible data — data that has not already been processed by application-layer encryption or compression. When an application performs deduplication, compression, or encryption before data reaches the array, PowerStore's inline engine achieves significantly lower ratios.

Three families of application-level reduction were identified (see research page phase-14-app-level-drr-variants.md):

  1. Application compression (Oracle HCC, SQL Server Page Compression, MongoDB WiredTiger) — removes the block-level redundancy that array dedup targets.
  2. Encryption (Oracle TDE, SQL Server TDE, pgcrypto, LUKS) — randomises the bitstream, making blocks incompressible and appearing unique. Dell KB000267460 classifies host-encrypted data as "unreducible".
  3. Backup agent source-side dedup (Veeam with compression+dedup enabled, Commvault, DDVE) — exhausts reduction opportunities before data reaches the array.

Decision

Model these scenarios as additional subcategory rows in DRR.csv rather than adding code logic to the classification engine or calculation layer.

The existing (category, subcategory) → DRR float lookup in DRRTable already provides the necessary flexibility. New rows such as:

Database;Oracle - TDE (Encrypted);1.5
Database;Oracle - HCC + TDE;1.2
VM Replication;Data Domain Virtual Edition (DDVE);1.0

…appear automatically in the workload dropdown on the review page without any UI code changes. Pre-sales engineers can manually select the appropriate variant when they know a VM uses encryption or source-side compression.

Companion classifier rules (priority 88–97) match common naming conventions (e.g. ORACLE-TDE-01, SQL-PAGE-DB) to auto-assign variants for recognisable names. Classifier patterns use regex lookaheads for combined scenarios (e.g. a VM name containing both "ORACLE", "HCC", and "TDE" → "Oracle - HCC + TDE").

Consequences

  • DRR.csv is the single source of truth for all DRR values, including application-level degradation scenarios — no ratios are hardcoded anywhere else.
  • Adding a new scenario (e.g. "Oracle TDE + ASM") requires only a new CSV row and, optionally, a new classifier rule — no code changes to pipeline or UI.
  • Automatic detection via VM names is heuristic: many encrypted VMs will not have "TDE" or "ENC" in their name. Manual selection via the dropdown remains the primary workflow for confirmed encrypted workloads.
  • The conservative (lowest) DRR principle (ADR-005) still applies when a VM has multiple workloads assigned, including encrypted variants.