LILAM

📊 LILAM Status Quo: Performance & Stress-Test Report

1. Hardware Specifications (Host)

2. Software Stack Setup

3. Database Resources & Limits (Status Quo)

4. Resource Efficiency of the Worker Architecture

The analysis of resource consumption during the high-load phase demonstrates the scalability of the chosen architecture:

5. Throughput & Test Scenario

Scenario Objectives:

  1. Determine latencies between log timestamp and physical commit in LILAM.
  2. Measure performance during sequential write operations and parallel API events.
  3. Validate data consistency under extreme Mixed Workload.

Test Pair Setup

Pair Mode Throughput (EPS) Status
Pair 1 (Bulk) FORALL ~1,300 - 1,900 Stable over 4M records
Pair 2 (Latency) Row-by-Row ~460 - 800 Drop at ~450k (I/O saturation)

6. Latency Visualization (Real-Time Capability)

Comparison of timestamps (Producer vs. Engine Entry).

Metric Value (Clean Run 100k) Value (Heavy Load 4M)
Average (Avg) 4.14 ms ~47,000 ms (Congestion phase)
Maximum (Max) 2,037.96 ms > 60,000 ms
Jitter (StdDev) 510.19 Extremely high (CPU saturation)

📊 System Metrics (Post-Mortem Analysis)

Cumulative statistics of the Oracle instance after completion of the test series.

A. Resource Saturation (Wait Events)

Event Name Total Waits Total Time (Sec.) Cause
library cache: mutex X 282,799 1,065.85 CPU thread limit (23ai Free)
library cache pin 602,926 488.94 Concurrent PL/SQL access
resmgr:cpu quantum 4,256 623.81 Throttling by Resource Manager
log file sync 116 2.48 SSD acknowledgment (Commit)

B. Data Throughput (I/O Statistics)

Metric Value
Redo Size 4,499.18 MB
Physical Writes 99.10 MB
Physical Reads 717.18 MB

C. Latency Distribution (Mixed Workload Analysis)

Latency Class Share (%) Evaluation
< 5 ms (Real-Time) 52.04 % System baseline speed
5 - 100 ms (Good) 12.52 % Minor resource conflicts
> 100 ms (Delayed) 35.44 % Hardware saturation (I/O & threads)

Summary Evaluation: The metrics confirm 100% utilization of the Oracle Free Edition. In particular, the high values in the library cache area clarify that the software architecture (LILAM) reached the physical management limits of the database instance. The system remained consistent at all times despite this massive saturation.

The LILAM engine demonstrates high algorithmic efficiency. Validation tests with and without an active rule set (7 rules via associative array) showed identical throughput rates. This proves that the logical processing layer generates negligible overhead compared to the physical I/O throughput of the hardware.