ADS (SAP) vs OpenADS — heavy engine bench

This is the primary compatibility/performance test: same Harbour workload, same data directory, two engines swapped at runtime:

Engine	DLL	Role
SAP Advantage	ace64.dll (+ adsloc64.dll local)	Reference / production baseline
OpenADS	openace64.dll	Drop-in replacement under test

The light NAV SQL backends (sqlite / postgresql / odbc-remote) measure AdsOpenTable over external databases. This bench measures the native DBF/CDX engine path — SEEK, AOF, SQL-on-DBF, append/concurrency — where OpenADS must match SAP ADS behaviour and throughput.

Scope (heavy)

Workload	What it proves
seek_eq	CDX index seek hot path (200× dbSeek)
aof_eq	Rushmore-style filter walk (AdsSetAOF)
rdd_scan_eq	Full table scan baseline
sql_count / sql_fetch	SQL engine on .dbf (OpenADS AdsExecuteSQLDirect)
openads_stress	Generate 100k+ row synthetic DBF + indexes
openads_concurrency_stress	Multi-thread append/lock/read on one table

The harness here is an A/B runner: it runs the same SEEK/AOF/SQL suite twice, once with openace64.dll and once with SAP ace64.dll beside the exe, and diffs the medians per workload.

Measured cycle

1. Prepare — openads_bench or openads_stress creates bench.dbf + bench.cdx (deterministic seed; default 10k–100k rows).
2. Run A — copy SAP ace64.dll (+ adsloc64.dll) → execute workloads → JSON/log.
3. Run B — copy OpenADS openace64.dll → same data dir → same workloads.
4. Compare — median ms per workload; flag regressions > threshold (e.g. 1.5×).

No PII. Synthetic data only. Portable paths via env (no hardcoded drive letters).

Environment variables

Variable	Purpose
OPENADS_ADS_SAP_SDK_DIR	Advantage SDK support files (collate, icu .dat, cfg)
HB_INSTALL / OPENADS_HB_BIN	Harbour tree (ace64.lib in lib/ for hbmk2 -lrddads -lace64)
OPENADS_ADS_SAP_DLL_DIR	Folder with SAP ace64.dll runtime (FWH samples/misc or Advantage install)
OPENADS_ADS_OPEN_DLL_DIR	Folder with openace64.dll
OPENADS_ADS_DATA_DIR	Shared DBF directory (default tools/bench/fixtures/ads_compare_data)
OPENADS_ADS_ROWS	Row count (default 10000)
OPENADS_ADS_REPEATS	Median repeats (default 3)
OPENADS_HB_BIN	hbmk2.exe for Harbour build

Harbour / SAP link: <harbour>/lib/ace64.lib (contrib/rddads — same as hbmk2 -lace64). SAP runtime DLL: ace64.dll beside the exe (typical: FiveWin FWH*/samples/misc, or your Advantage redistribute). The Advantage SDK supplies collate/icu cfg and ace32 only — not ace64.dll. OpenADS: openace64.dll / openace64.lib from the OpenADS build.

How to run

pwsh tools/scripts/prepare_ads_compare_data.ps1 -Rows 2000 -Build
pwsh tools/scripts/run_ads_vs_openads_bench.ps1 -Rows 2000 -Repeats 3

Data is created with SAP ace64 (ads_prepare_sap_64.exe). Each engine builds its own bench.cdx (CDX formats are not interchangeable). RDD workloads run by default; SQL-on-DBF runs on OpenADS only (SAP local SQL rejects bench.dbf link syntax).

Produces:

• tools/bench/results/ads_vs_openads_latest.json
• tools/bench/results/OPENADS_ADS_VS_OPENADS_FWH.pdf

Success criteria

• Both engines complete all workloads without ABI errors.
• OpenADS within agreed ratio of SAP ADS per workload (project-specific SLA).
• Results committed on the PR branch when the SAP DLL path is available in CI/dev.

Relation to NAV bench

Suite	Path	Backend
NAV bench (PR #26)	demo_nav_multidb.prg	Remote SQL URIs (sqlite/pg/odbc)
ADS vs OpenADS	ads_compare engine A/B + stress tools	Local DBF/CDX engine

Both are required for a complete OpenADS Plus + engine validation story.