Varpulis vs Apama Performance Analysis

For comprehensive benchmark results, see benchmarks-apama-comparison.md — 7 scenarios, MQTT connector + CLI ramdisk modes, memory + throughput comparisons.

Current Benchmark Results (2026-02-09)

Measured across 7 CEP scenarios with 100K events, median of 3 runs:

Mode	Varpulis	Apama	Varpulis Memory	Apama Memory
MQTT Connector	6–10K/s	6K/s	10–57 MB	85–153 MB
CLI Ramdisk	234–335K/s	195–221K/s	36–66 MB	166–190 MB

Varpulis wins 4 of 5 CPU-bound scenarios (1.2–1.3x) and uses 2x–16x less memory in every scenario.

Optimization History

PERF-04 (2026-01-31): Changed event_sources: HashMap<String, Vec<String>> to Arc<[String]>. Arc clone is O(1) atomic increment vs O(n) deep copy. Improvement: ~4-6%.

PERF-05 (2026-02-07): Sync processing path, zero-clone preload, batch size increase (1K→10K), zero-alloc field splitting. Improvement: ~40% (140K→234K evt/s on filter benchmark).

Optimization Analysis

Resolved Bottlenecks

The following bottlenecks from the January 2026 analysis have been fixed:

1. Stream name Vec cloning — Resolved: Changed to Arc<[String]> (PERF-04)
2. Event file parsing — Resolved: Zero-alloc split_fields() returns Vec<&str> (PERF-05)
3. Batch processing — Resolved: Batch size 10K in preload mode, sync path (PERF-05)
4. Event cloning — Resolved: Zero-clone preload with into_iter() + drain() (PERF-05)

Remaining Optimization Opportunities

1. Object pooling for events: Pre-allocated pools for high-throughput scenarios
2. SmallVec for small collections: Avoid heap for collections < 8 items
3. SIMD-accelerated field parsing: Use memchr for delimiter scanning

Test Coverage Status

Current test coverage (COV-01/COV-02 - COMPLETE):

Total: 752+ tests passing across all crates

Module	Tests	Status
varpulis-cli	114 tests	✅
varpulis-core	57 tests	✅
varpulis-parser	77 tests	✅
varpulis-runtime	358 tests	✅
integration tests	73 tests	✅
e2e tests	49 tests	✅

Runtime Module Coverage

Module	Tests	Coverage
engine/mod.rs	~40 tests	✅ Comprehensive
event.rs	17 tests	✅ Good
event_file.rs	26 tests	✅ Good
simulator.rs	12 tests	✅ Good
sase.rs	~25 tests	✅ Good
window.rs	~20 tests	✅ Good
aggregation.rs	~15 tests	✅ Good
pattern.rs	40+ tests	✅ Comprehensive
(removed modules)	-	-

Pattern Tests Coverage (40+ tests)

• Event matching with filters (eq, neq, gt, ge, lt, le)
• Logical operators (and, or, xor, not)
• Sequence patterns (followed_by, within)
• PatternContext merging
• Engine state management