benchmarks

Argus Ring Buffer Performance Benchmarks

This directory contains isolated performance benchmarks for the BoreasLite ring buffer implementation used in Argus. The benchmarks are separated from the main test suite to provide accurate performance measurements without interference from intensive unit tests.

Benchmark Results

All benchmarks were executed on AMD Ryzen 5 7520U.

Single Event Processing (Optimized Path)

BenchmarkBoreasLite_SingleEvent-8    47011258    25.63 ns/op    39.02 Mops/sec    0 B/op    0 allocs/op

• Latency: 25.63 nanoseconds per operation
• Throughput: 39.02 million operations per second
• Memory: Zero allocations in hot path

Write Operations

BenchmarkBoreasLite_WriteFileEvent-8    67764067    53.20 ns/op    18.80 Mops/sec    0 B/op    0 allocs/op

• Latency: 53.20 nanoseconds per operation
• Throughput: 18.80 million operations per second
• Memory: Zero allocations

Multi-Producer Single Consumer (MPSC)

BenchmarkBoreasLite_MPSC-8    31231618    34.77 ns/op    28.76 Mops/sec    0 B/op    0 allocs/op

• Latency: 34.77 nanoseconds per operation under concurrent load
• Throughput: 28.76 million operations per second
• Scalability: Performance maintained across multiple producers

Comparison with Go Channels

BoreasLite

BenchmarkBoreasLite_vsChannels/BoreasLite-8    23317405    45.72 ns/op    21.87 Mops/sec    0 B/op    0 allocs/op

Go Channels

BenchmarkBoreasLite_vsChannels/GoChannels-8    18621222    61.43 ns/op    16.28 Mops/sec    0 B/op    0 allocs/op

Performance Delta

• BoreasLite: 21.87 million ops/sec
• Go Channels: 16.28 million ops/sec
• Improvement: 34.3% faster than native Go channels

High Throughput Sustained Load

BenchmarkBoreasLite_HighThroughput-8    27012850    53.88 ns/op    18.56 Mops/sec    0 B/op    0 allocs/op

• Sustained throughput: 18.56 million operations per second
• Buffer size: 8192 events
• Strategy: Large batch optimization

Technical Implementation

Ring Buffer Architecture

• Type: Multiple Producer Single Consumer (MPSC)
• Synchronization: Lock-free atomic operations
• Memory Layout: Cache-line aligned, power-of-2 sizing
• Event Size: 128 bytes (2 cache lines)

Optimization Strategies

1. SingleEvent: Ultra-low latency for 1-2 files (25ns)
2. SmallBatch: Balanced performance for 3-20 files
3. LargeBatch: High throughput for 20+ files with 4x unrolling

Memory Characteristics

• Zero allocations in all hot paths
• Fixed memory footprint: 8KB + (128 bytes × buffer_size)
• Cache efficiency: Power-of-2 ring buffer with atomic sequence numbers

Running Benchmarks

Execute all benchmarks:

go test -bench="BenchmarkBoreasLite.*" -run=^$ -benchmem

Execute specific benchmark:

go test -bench=BenchmarkBoreasLite_SingleEvent -run=^$ -benchmem

Execute with multiple iterations:

go test -bench="BenchmarkBoreasLite.*" -run=^$ -benchmem -count=3

Dependencies

• github.com/agilira/argus: Main library (via replace directive)
• github.com/agilira/go-timecache: High-performance timestamp caching

Notes

• Benchmarks use minimal processing functions to isolate ring buffer performance
• All measurements include complete write-to-process cycles where applicable
• MPSC benchmarks use GOMAXPROCS concurrent producers
• Results represent sustainable performance under continuous load