Performance Benchmarks

Fresh

Grounded performance metrics from real-world ontologies (PROV-O, RDF+RDFS, SHACL) with binary-search methodology for exact limits.

Testing Methodology

Environment:

• Hardware: MacBook Pro M1
• Node.js: v24.7.0
• Method: Median of multiple runs, warmed-up parser
• Data: Synthetic and real-world ontologies

Test Types:

1. Synthetic Scaling — Precise quad count testing
2. Real-World Ontologies — PROV-O, RDF+RDFS, SHACL
3. Binary Search — Exact limit determination
4. Conservative Limits — 95% confidence with 10% safety margin

Key Metrics

Metric	Value	Use Case
60fps Frame	4,527 quads	Interactive knowledge graphs
1-Second Batch	225,059 quads	Background reindexing, imports
Sustained Rate	252K quads/sec	Continuous processing

Real-World Efficiency

Document Type	Quads	Size (KB)	Quads/sec
PROV-O	944	141	27,505
RDF+RDFS	231	39	36,903
SHACL	286	109	23,692
Mixed Set	1,461	288	51,684

Scaling Characteristics

60fps Performance

Scale	Quads	Parse Time	Status
Conservative Limit	4,527	16.12ms	within 96.7% budget
Maximum Limit	5,031	19.15ms	exceeds 16.67ms budget

1-Second Performance

Scale	Quads	Parse Time	Status
Conservative Limit	225,059	890ms	within 89.0% budget
Maximum Limit	250,066	990ms	within budget

Enterprise Scaling

Document Collection Limits

Enterprise Size	Total Quads	Parse Time	Architecture
Small	7,602	147ms	Full reparse OK
Medium	28,670	647ms	Background reparse
Large	72,830	1,461ms	Incremental only

Documents Per Second

Document Type	Parses/sec	Quads/sec
PROV-O	60	27,505
RDF+RDFS	349	36,903
SHACL	153	23,692
Mixed Set	35	51,684

Binary Search Results

60fps Target: 16.67ms
├── Conservative: 4,527 quads @ 16.12ms (96.7% budget)
└── Maximum: 5,031 quads @ 19.15ms (exceeds budget)

1-Second Target: 1000ms
├── Conservative: 225,059 quads @ 890ms (89.0% budget)
└── Maximum: 250,066 quads @ 990ms (within budget)

Performance Optimization

Character-Based Tokenization

Improvement: 20-28% faster than regex-based approaches

Techniques:

• Direct character inspection
• No regex engine overhead
• Predictable O(n) complexity
• Memory-efficient state tracking

Memory Management

• ~640 bytes per quad after GC
• Streaming-friendly single-pass
• No full document copies
• Efficient indexing structures

Incremental Updates

• Under 5ms per ontology addition
• O(new) complexity
• Real-time capable for under 4K quads
• Background processing for larger sets

Usage Recommendations

Real-Time Applications (60fps)

• Document size: max 4K quads
• Update pattern: Incremental only
• Use case: Interactive knowledge graphs
• Architecture: Pre-built indexes, background indexing for larger sets

Batch Processing (1-Second)

• Document size: max 225K quads
• Update pattern: Full reparse
• Use case: Background reindexing, imports
• Architecture: Worker thread processing, chunked processing for large files

Enterprise Knowledge Bases

Scale Guidelines:

Size	Quads	Approach
Small	under 4K	Real-time updates
Medium	4K to 225K	Batch reindexing
Large	over 225K	Streaming architecture

Bottom Line

MD-LD delivers enterprise-scale performance with 252K quads/sec sustained throughput and real-time capabilities for interactive applications.

Key takeaway: MD-LD can reliably process 4K quads at 60fps and 225K quads per second, making it suitable for both real-time interactive applications and large-scale knowledge management systems.