Pattern Comparison Report: Streaming Query Approaches

Date: February 4, 2026
Experiment: Pattern-Based Accuracy and Performance Comparison
Duration: 120 seconds per test, 4Hz sampling (480 observations each)

Executive Summary

This report presents a comprehensive comparison of three streaming query processing approaches across five distinct data patterns. Each pattern tests different characteristics of the streaming data to evaluate robustness, accuracy, and resource efficiency.

Tested Approaches

Fetching (Client-Side) - Baseline approach with client-side aggregation
Approximation - Rate-based sub-query approach
Chunked - Incremental aggregation with GCD-based sub-queries

Data Patterns Tested

Low Variability - Gaussian noise (μ=-23.0, σ=0.25)
Step Pattern - Step change from -23.0 to -15.0 at t=60s
Spike Pattern - Brief spike from -23.0 to -5.0 for 1.25s
Low Frequency Oscillation - Sinusoidal (μ=-23.0, A=5.0, f=0.05Hz)
High Frequency Oscillation - Sinusoidal (μ=-23.0, A=3.0, f=0.5Hz)

Overall Results Summary

Accuracy Comparison

Pattern	Approach	Result Value	MAPE vs Fetching (%)	Absolute Error
low_variability	Fetching	-22.987	0.00	0.000
low_variability	Approximation	-22.993	-0.03	0.006
step_pattern	Fetching	-17.684	0.00	0.000
step_pattern	Approximation	-17.032	-3.69	0.652
spike_pattern	Fetching	-22.930	0.00	0.000
spike_pattern	Approximation	-22.741	-0.82	0.189
low_freq_oscillation	Fetching	-23.002	0.00	0.000
low_freq_oscillation	Approximation	-23.003	-0.01	0.002
high_freq_oscillation	Fetching	-23.000	0.00	0.000
high_freq_oscillation	Approximation	-22.989	-0.05	0.011
low_variability	Chunked	-22.987	0.00	0.000
step_pattern	Chunked	-19.441	9.94	1.758
spike_pattern	Chunked	-22.870	0.26	0.060
low_freq_oscillation	Chunked	-22.886	0.50	0.116
high_freq_oscillation	Chunked	-22.994	0.03	0.006

Resource Usage Comparison

Pattern	Approach	Avg CPU User	Avg CPU System	Avg Memory (MB)	Peak Memory (MB)
low_variability	Fetching	3936.61	1149.97	45.23	45.58
low_variability	Approximation	4523.14	1208.91	47.43	42.32
step_pattern	Fetching	4003.88	1162.75	45.35	41.23
step_pattern	Approximation	4200.70	1146.08	48.66	41.98
spike_pattern	Fetching	3944.82	1091.13	44.23	39.06
spike_pattern	Approximation	4548.93	1167.90	47.12	42.16
low_freq_oscillation	Fetching	3912.92	1097.71	45.52	42.74
low_freq_oscillation	Approximation	4261.94	1160.81	48.69	42.14
high_freq_oscillation	Fetching	4023.16	1192.15	45.50	45.43
high_freq_oscillation	Approximation	4398.86	1216.76	48.36	42.26
low_variability	Chunked	4507.48	1261.40	48.23	42.15
step_pattern	Chunked	4587.61	1227.81	48.28	42.27
spike_pattern	Chunked	4383.96	1168.35	48.64	42.31
low_freq_oscillation	Chunked	4763.17	1256.17	47.30	41.95
high_freq_oscillation	Chunked	4476.21	1213.59	48.45	42.32

Detailed Analysis by Pattern

Low Variability

Pattern Characteristics:

Type: Gaussian noise
Mean (μ): -23.0
Standard Deviation (σ): 0.25
Expected Behavior: Stable mean with minimal variance

Results:

Approach	Windows	Result Value	MAPE (%)	Absolute Error	CPU User	CPU System	Memory (MB)	Peak Mem (MB)
Fetching	3	-22.987	0.00	0.000	3936.6	1150.0	45.2	45.6
Approximation	2	-22.993	-0.03	0.006	4523.1	1208.9	47.4	42.3
Chunked	2	-22.987	0.00	0.000	4507.5	1261.4	48.2	42.1

Analysis:

This pattern tests the approaches' ability to handle low-variance, steady-state data
Expected: All approaches should produce highly accurate results due to data stability

Step Pattern

Pattern Characteristics:

Type: Step function
Initial Value (v₁): -23.0
Final Value (v₂): -15.0
Step Time (t_step): 60s
Expected Behavior: Abrupt transition halfway through observation period

Results:

Approach	Windows	Result Value	MAPE (%)	Absolute Error	CPU User	CPU System	Memory (MB)	Peak Mem (MB)
Fetching	2	-17.684	0.00	0.000	4003.9	1162.8	45.4	41.2
Approximation	2	-17.032	-3.69	0.652	4200.7	1146.1	48.7	42.0
Chunked	2	-19.441	9.94	1.758	4587.6	1227.8	48.3	42.3

Analysis:

This pattern tests handling of abrupt regime changes
The 120s window with 60s step creates a 50-50 split between low and high values
Expected mean: (-23.0 + -15.0) / 2 = -19.0

Spike Pattern

Pattern Characteristics:

Type: Transient spike
Base Value (v_base): -23.0
Spike Value (v_spike): -5.0
Spike Duration (Δt): 1.25s
Spike Position: Center of observation window (t=60s)
Expected Behavior: Brief high-magnitude deviation

Results:

Approach	Windows	Result Value	MAPE (%)	Absolute Error	CPU User	CPU System	Memory (MB)	Peak Mem (MB)
Fetching	2	-22.930	0.00	0.000	3944.8	1091.1	44.2	39.1
Approximation	2	-22.741	-0.82	0.189	4548.9	1167.9	47.1	42.2
Chunked	2	-22.870	0.26	0.060	4384.0	1168.3	48.6	42.3

Analysis:

This pattern tests handling of transient anomalies
Spike duration: 1.25s out of 120s total (1.04% of window)
Approximation approaches may smooth out or miss brief spikes
Tests temporal resolution and anomaly preservation

Low Freq Oscillation

Pattern Characteristics:

Type: Sinusoidal oscillation
Mean (μ): -23.0
Amplitude (A): 5.0
Frequency (f): 0.05 Hz
Sampling Rate: 4 Hz (80 samples per cycle)
Expected Behavior: Slow, smooth oscillation with 6 complete cycles in 120s

Results:

Approach	Windows	Result Value	MAPE (%)	Absolute Error	CPU User	CPU System	Memory (MB)	Peak Mem (MB)
Fetching	2	-23.002	0.00	0.000	3912.9	1097.7	45.5	42.7
Approximation	2	-23.003	-0.01	0.002	4261.9	1160.8	48.7	42.1
Chunked	2	-22.886	0.50	0.116	4763.2	1256.2	47.3	42.0

Analysis:

Tests ability to capture slow periodic variations
80 samples per cycle provides excellent temporal resolution
All approaches should accurately capture this pattern
Expected mean over full cycles: μ = -23.0

High Freq Oscillation

Pattern Characteristics:

Type: Sinusoidal oscillation
Mean (μ): -23.0
Amplitude (A): 3.0
Frequency (f): 0.5 Hz
Sampling Rate: 4 Hz (8 samples per cycle)
Expected Behavior: Rapid oscillation with 60 complete cycles in 120s

Results:

Approach	Windows	Result Value	MAPE (%)	Absolute Error	CPU User	CPU System	Memory (MB)	Peak Mem (MB)
Fetching	2	-23.000	0.00	0.000	4023.2	1192.2	45.5	45.4
Approximation	2	-22.989	-0.05	0.011	4398.9	1216.8	48.4	42.3
Chunked	2	-22.994	0.03	0.006	4476.2	1213.6	48.5	42.3

Analysis:

Tests handling of higher frequency periodic patterns
8 samples per cycle is adequate but less margin than low-frequency case
Frequency is below Nyquist limit (2 Hz), so no aliasing expected
Tests aggregation accuracy with rapid fluctuations

Cross-Pattern Insights

Accuracy Trends

TBD after experiment completion

Resource Efficiency

TBD after experiment completion

Best Use Cases

Low Variability: All approaches suitable, choose based on resource constraints
Step Pattern: Tests boundary handling and regime change detection
Spike Pattern: Highlights temporal resolution differences
Low Freq Oscillation: Ideal for all approaches, well-sampled
High Freq Oscillation: Tests aggregation fidelity under rapid changes

Methodology

Query Configuration

Window Size: RANGE 120s (rolling window)
Slide Interval: STEP 60s (50% overlap)
Chunked Sub-queries: RANGE 30s, STEP 30s (GCD-based decomposition)

Data Configuration

Duration: 120 seconds
Sampling Rate: 4 Hz (250ms intervals)
Data Points: 480 observations per stream
Streams: Two correlated sensors (smartphoneX, wearableX)

Metrics Collected

Accuracy: MAPE (Mean Absolute Percentage Error) vs Fetching baseline
Latency: Time from query registration to result emission
CPU Usage: User and system CPU time (average)
Memory: Average and peak memory consumption (MB)

RSP-QL Query

PREFIX saref: <https://saref.etsi.org/core/>
PREFIX dahccsensors: <https://dahcc.idlab.ugent.be/Homelab/SensorsAndActuators/>

REGISTER RStream <output> AS
SELECT (AVG(?v) AS ?averageValue)
FROM NAMED WINDOW :w1 ON STREAM <mqtt://localhost:1883/smartphoneX> [RANGE PT120S STEP PT60S]
FROM NAMED WINDOW :w2 ON STREAM <mqtt://localhost:1883/wearableX> [RANGE PT120S STEP PT60S]
WHERE {
  WINDOW :w1 { ?obs saref:relatesToProperty dahccsensors:smartphoneX ; saref:hasValue ?v1 . }
  WINDOW :w2 { ?obs saref:relatesToProperty dahccsensors:wearableX ; saref:hasValue ?v2 . }
  BIND((?v1 + ?v2) / 2 AS ?v)
}

Appendix

Pattern Parameters Summary

Pattern	Key Parameters	Description
Low Variability	μ=-23.0, σ=0.25	Gaussian noise, minimal variation
Step Pattern	v₁=-23.0, v₂=-15.0, t_step=60s	Abrupt regime change at midpoint
Spike Pattern	v_base=-23.0, v_spike=-5.0, Δt=1.25s	Brief transient anomaly
Low Freq Osc.	μ=-23.0, A=5.0, f=0.05Hz	Slow periodic variation (6 cycles)
High Freq Osc.	μ=-23.0, A=3.0, f=0.5Hz	Rapid periodic variation (60 cycles)

Experiment Environment

OS: macOS
Node.js: v14+
MQTT Broker: Mosquitto (localhost:1883)
Experiment Duration: ~45 minutes (15 tests × 180s each)

Report generated automatically from experimental data. For questions or additional analysis, see experiment logs in pattern_comparison_results/

FilesExpand file tree

PATTERN_COMPARISON_REPORT.md

Latest commit

History

PATTERN_COMPARISON_REPORT.md

File metadata and controls

Pattern Comparison Report: Streaming Query Approaches

Executive Summary

Tested Approaches

Data Patterns Tested

Overall Results Summary

Accuracy Comparison

Resource Usage Comparison

Detailed Analysis by Pattern

Low Variability

Step Pattern

Spike Pattern

Low Freq Oscillation

High Freq Oscillation

Cross-Pattern Insights

Accuracy Trends

Resource Efficiency

Best Use Cases

Methodology

Query Configuration

Data Configuration

Metrics Collected

RSP-QL Query

Appendix

Pattern Parameters Summary

Experiment Environment