Performance Monitoring

Overview

Performance monitoring in MixModeler provides real-time visibility into how efficiently your models are running and which acceleration methods are being used. Understanding performance metrics helps you optimize your workflow, identify bottlenecks, and make informed decisions about hardware upgrades or model simplification.

Performance Indicators

Acceleration Badges

Located in the top-right toolbar, acceleration badges show active performance enhancements:

🚀 WASM Badge (Always Present):

• Indicates WebAssembly acceleration active

• Hover to see WASM engine version

• Green color confirms optimal performance

🖥️ GPU Badge (When Available):

• Indicates GPU acceleration active

• Hover to see GPU model and utilization

• Click for detailed GPU performance metrics

• Green = active, Yellow = limited, Absent = unavailable

Operation Timing Display

After major operations, MixModeler displays performance information:

Bottom-Right Notification:

✅ Model fitted in 0.73s using WASM

Information Shown:

• Operation completed (checkmark indicates success)

• Duration in seconds

• Acceleration method used (WASM, WebGPU, or CPU)

Interpretation:

• <1 second: Excellent performance

• 1-5 seconds: Good performance

• 5-15 seconds: Acceptable for complex operations

• 15 seconds: Consider optimization

Console Performance Logs

For detailed analysis, browser console shows comprehensive timing:

Example Log Output:

🚀 Using WASM for model fitting...
   Data preparation: 45ms
   Matrix operations: 580ms
   Result processing: 92ms
✅ WASM fitting completed in 717ms total

Access Console:

• Chrome/Edge: F12 or Ctrl+Shift+J

• Firefox: F12 or Ctrl+Shift+K

• Safari: Cmd+Option+C

Performance Metrics

Operation Duration

Time from operation start to completion.

What It Measures: Total wall-clock time including all processing steps

Benchmarks by Operation:

Operation	Small Dataset	Medium Dataset	Large Dataset
Data Upload	<0.5s	0.5-2s	2-5s
Model Fitting (OLS)	0.3-0.7s	0.7-2s	2-8s
Model Fitting (Bayesian)	60-120s	120-300s	300-600s
Diagnostics Suite	0.3-0.6s	0.6-1.5s	1.5-4s
Correlation Matrix	0.2-0.5s	0.5-1.2s	1.2-3s
Variable Testing	0.5-1.5s	1.5-4s	4-10s
Decomposition Analysis	0.4-0.8s	0.8-2s	2-5s

Small: <30 vars, <100 obs | Medium: 30-100 vars, 100-300 obs | Large: 100+ vars, 300+ obs

Acceleration Method

Which technology powered the operation.

Methods:

• WebGPU: Highest performance, GPU-accelerated

• WASM: Good performance, always available

• CPU/JavaScript: Baseline, fallback method

Priority: MixModeler tries WebGPU first, then WASM, then CPU

Typical Distribution:

• With GPU: 70% WebGPU, 25% WASM, 5% CPU

• Without GPU: 0% WebGPU, 85% WASM, 15% CPU

Speedup Factor

How much faster accelerated methods are compared to baseline.

Calculation: Baseline Time / Accelerated Time

Example:

• CPU time (estimated): 4.2s

• WASM time (actual): 0.7s

• Speedup: 6x faster

Typical Speedups:

• WASM vs CPU: 5-10x

• GPU vs CPU: 15-60x (when available)

• GPU vs WASM: 3-8x (when both available)

Memory Usage

RAM consumed during operations (particularly relevant for Bayesian modeling).

Display: Shows peak memory during operation

Benchmarks:

• Small OLS Model: 50-150 MB

• Medium OLS Model: 150-400 MB

• Large OLS Model: 400-1000 MB

• Bayesian MCMC (4 chains, 2000 draws): 500-2000 MB

Concern Thresholds:

• <1 GB: No concerns

• 1-2 GB: Monitor if multiple tabs open

• 2-4 GB: Close other applications

• 4 GB: Consider reducing model complexity

Detailed Performance View

Accessing Performance Dashboard

1. Click acceleration badge (WASM or GPU) in toolbar

2. Select "Performance Details" from dropdown

3. View comprehensive performance breakdown

Alternative Access: Settings → Performance → View Detailed Metrics

Dashboard Components

Recent Operations Table:

Operation	Duration	Method	Speedup	Timestamp
Model Fit	0.68s	WASM	6.2x	14:23:45
Diagnostics	0.41s	WASM	7.8x	14:23:46
Correlation	0.15s	WebGPU	23.1x	14:24:02
Variable Test	1.24s	WASM	6.8x	14:24:15

Session Summary:

• Total operations: 47

• Total time saved: 3 minutes 24 seconds

• Average speedup: 8.3x

• Primary method: WASM (68%), WebGPU (32%)

GPU Utilization (when available):

• Current usage: 34%

• Peak usage: 78%

• Average usage: 42%

• Total GPU time: 8.3 seconds

System Information:

• Browser: Chrome 120.0

• WASM version: 1.0

• GPU: NVIDIA RTX 3060 (detected)

• Available RAM: 14.2 GB / 16 GB

Performance Optimization

Identifying Bottlenecks

Slow Data Upload (>5s for medium dataset):

• Possible Cause: Large file, slow disk, browser cache issues

• Check: File size, number of variables

• Solution: Reduce variables, clear cache, use faster storage

Slow Model Fitting (>10s for OLS):

• Possible Cause: Too many variables, multicollinearity, no acceleration

• Check: Variable count, acceleration badge status

• Solution: Remove correlated variables, enable GPU, simplify model

Slow Diagnostics (>5s):

• Possible Cause: Many diagnostic tests, large dataset

• Check: Number of tests enabled, dataset size

• Solution: Run only essential tests initially

Slow Bayesian MCMC (>10 minutes):

• Possible Cause: Too many draws, poor convergence, complex model

• Check: MCMC settings, convergence diagnostics

• Solution: Use Fast Inference mode, reduce draws initially, simplify model

Optimization Strategies

For Faster Iterations:

1. Start with subset of variables (10-20)

2. Use OLS before Bayesian

3. Enable Fast Inference for Bayesian exploration

4. Reduce diagnostic frequency during development

5. Leverage GPU if available

For Large Datasets:

1. Ensure GPU acceleration active

2. Close unnecessary browser tabs

3. Process in batches if needed

4. Use standardized variables (improves numerical stability)

5. Consider data reduction techniques

For Bayesian Models:

1. Use Fast Inference (SVI) for initial exploration

2. Start with fewer chains (2) and draws (1000)

3. Increase gradually only if convergence poor

4. Monitor memory usage

5. Switch to full MCMC only for final model

For Memory-Constrained Systems:

1. Close other applications

2. Use single browser tab

3. Reduce Bayesian chains and draws

4. Process variables in groups

5. Clear browser cache regularly

Benchmarking Your System

Running a Standard Benchmark

To understand your system's baseline performance:

1. Load the demo dataset (50 variables, 104 observations)

2. Build model with all variables

3. Run OLS model

4. Note timing displayed

5. Run full diagnostics suite

6. Note timing displayed

7. Generate correlation matrix

8. Note timing displayed

Interpreting Benchmark Results

High Performance (with GPU):

• Model fitting: <0.5s

• Diagnostics: <0.3s

• Correlation: <0.1s

• Total workflow: <1.5s

Good Performance (WASM only):

• Model fitting: 0.5-1.0s

• Diagnostics: 0.3-0.6s

• Correlation: 0.2-0.4s

• Total workflow: 1.5-3s

Adequate Performance (older hardware):

• Model fitting: 1.0-2.0s

• Diagnostics: 0.6-1.2s

• Correlation: 0.4-0.8s

• Total workflow: 3-5s

Slow Performance (needs upgrade):

• Model fitting: >2s

• Diagnostics: >1.2s

• Correlation: >0.8s

• Total workflow: >5s

Comparing to Reference Systems

Budget Laptop (Intel i3, 8GB RAM, integrated graphics):

• WASM only

• Model fitting: ~1.5s

• Full workflow: ~4s

Mid-Range Laptop (Intel i5/AMD Ryzen 5, 16GB RAM, no dedicated GPU):

• WASM only

• Model fitting: ~0.8s

• Full workflow: ~2.5s

Gaming Laptop (Intel i7/AMD Ryzen 7, 16GB RAM, NVIDIA GTX 1660):

• WASM + GPU

• Model fitting: ~0.3s

• Full workflow: ~0.8s

Workstation (Intel i9/AMD Ryzen 9, 32GB RAM, NVIDIA RTX 3070):

• WASM + GPU

• Model fitting: ~0.2s

• Full workflow: ~0.5s

Mac M1/M2 (8-16GB unified memory):

• WASM + GPU (Metal)

• Model fitting: ~0.4s

• Full workflow: ~1.0s

Performance Troubleshooting

Diagnosis Flowchart

Is WASM badge present?

• No → Browser issue, try update/reinstall

• Yes → Proceed

Are operations taking >5s for medium models?

• No → Performance is normal

• Yes → Proceed

Is GPU badge present?

• No → GPU unavailable, expect WASM-only speeds

• Yes → GPU should be helping, proceed

Check console logs - are operations using GPU?

• Yes → GPU active, may need better GPU

• No → GPU fallback occurring, investigate why

Common causes of GPU fallback:

• Insufficient VRAM

• GPU busy with other tasks

• Driver compatibility issues

• Operation too small for GPU benefit

Quick Performance Fixes

Fix 1: Clear Browser Cache

• Often resolves slow loading and acceleration issues

• Chrome: Ctrl+Shift+Delete → Clear cached images and files

• Restart browser after clearing

Fix 2: Close Other Tabs

• Each tab consumes memory and may use GPU

• Close unused tabs before intensive operations

• Particularly important for Bayesian modeling

Fix 3: Update Graphics Drivers

• Outdated drivers limit GPU performance

• Visit GPU manufacturer website (NVIDIA, AMD, Intel)

• Download and install latest drivers

• Restart computer after installation

Fix 4: Enable Hardware Acceleration

• Chrome: Settings → System → Use hardware acceleration

• Ensure toggle is ON

• Restart browser

Fix 5: Restart Browser

• Memory leaks can slow performance over time

• Restart browser every few hours during heavy usage

• Particularly important during long modeling sessions

Export Performance Data

For Technical Support

If experiencing persistent performance issues:

1. Open Performance Dashboard

2. Click "Export Performance Report"

3. Saves JSON file with:

   • Operation timings

   • Acceleration methods used

   • System information

   • Error logs (if any)

Send to: [e-mail] with description of issues

For Internal Documentation

Track performance over time for capacity planning:

1. Export performance data monthly

2. Compare trends in operation timing

3. Identify if hardware upgrades needed

4. Document baseline vs current performance

Best Practices

Monitor Periodically: Glance at performance indicators occasionally, not constantly

Set Expectations: Know your system's baseline from benchmarking

Optimize Strategically: Focus optimization on actual bottlenecks, not all operations

Document Baselines: Record initial benchmark results for future comparison

Report Anomalies: If performance suddenly degrades, investigate immediately

Plan Upgrades: Use performance data to justify hardware investments

Educate Stakeholders: Share typical timing expectations to set realistic project timelines

Performance Impact on Workflow

Development Phase

With Good Performance (GPU + WASM):

• Rapid iteration (5-10 models per hour)

• Immediate feedback on changes

• Encourages experimentation

• Reduces fatigue and errors

With Poor Performance (CPU only):

• Slow iteration (1-2 models per hour)

• Waiting reduces focus

• Discourages exploration

• More likely to settle for suboptimal models

Time Multiplier: Good performance can make analysts 3-5x more productive

Production Phase

Impact on Deliverables:

• Faster final model validation

• Quick scenario analysis for stakeholders

• Responsive to last-minute changes

• Professional, efficient client interactions

Impact on Quality:

• More time for thorough testing

• Better explored model space

• Higher confidence in results

• More robust final recommendations

---

Next Steps: Explore Large Dataset Handling to optimize performance for models with hundreds of variables, or return to Advanced Features overview.