5 Stress Tests — continuous loop until housekeeping threshold

Each round = 25,500 signed ERP ops = 51 simulated business days at 500 tx/day. 102 days = 2 completed rounds with no degradation — the engine ran a quarter of a production year without a server. At threshold the engine housekeeps (seed compaction + op-log truncation), turns green, and continues.

50×Less Code28K vs 1.4M LOC
→
185×Fewer Tables5 vs 925 schemas
→
$0Server Costvs $1,000s/mo
→
—10K-op verifycryptographic chain
→
0Days simulatedcumulative all rounds
0 DAYS SIMULATED
Round — / ∞
0Records traversed
0.0sWall time
—Ops / sec
0Rounds done
⚙ HOUSEKEEPING — seed compacting · op-log truncating · born tenants preserved…
Ready — click to start continuous benchmark loop
① Session Memory
Heap growth over 5K ops — GC stable?
IDLE
↑—vs iDempiere
—Heap delta
—Leak?
What this checks: does the heap grow as it runs? Flat = no leak, GC stable. A leak test — not a memory-size claim.
vs iDempiere: ~512 MB JVM baseline before first user — our delta is — smaller
② Daily 10K + Backup
a day's 10,000 Sales Orders, then End-of-Day backup
IDLE
—orders/sec · KernelERP
—10K orders
—+ EOD backup
Same unit = 1 completed Sales Order (doc-no + C_Order + C_OrderLine + C_OrderTax + stock reservation). KernelERP ~2 s (in a browser tab) · iDempiere ~6–7 min for 10,000 — same box, ~170×. KernelERP also signs each order into a hash-chain (iDempiere has none) and writes the EOD backup. Both single-threaded; measured — iDempiere ≈ 35 ms/order over 300 real completeIt(), KernelERP folds the same rows. iDempiere's 35 ms also re-runs its Java rule engine (callouts/validators/tax); like-for-like on the persisted order, not on re-running that logic.
vs iDempiere: ~6–7 min for 10K — measured on this box over 300 real completeIt() Standard-Order completions (iDempiere 13, GardenWorld), ~35 ms/doc median; same box has no add-on backup.
③ Cold-Load vs Size
Hydration time as seed grows
IDLE
↑—vs iDempiere
—Baseline
—At 5× size
What this checks: open → ready (hydration) time as the seed database gets bigger.
vs iDempiere: ~2–5 min cold start (JVM + OSGi + PG pool) — our load: — faster
④ Multi-Peer Sync
N nodes merge — no server
IDLE
↑—vs iDempiere
—10-peer merge
—Scaling
What this checks: merge N offline copies into one. The edge is the network round-trips you skip — set it with the dial above.
vs iDempiere: server required — every peer op is a network round-trip (~50–200 ms each); 500 ops ≈ 25–100 s via server. Our merge: — faster, no server
⑤ verifyChain
Cryptographic chain at 100→10K ops
IDLE
↑∞no equiv. in iDempiere
—10K ops
—Ops/sec
What this checks: tamper-check speed at scale (our own engine). iDempiere has no built-in equivalent — not a head-to-head.
vs iDempiere: no cryptographic audit trail — tamper-evidence requires an external audit log add-on; this engine has it built in