Skip to content

Performance Benchmarks

We benchmarked the modern Sentinel M4/M8 (Seccomp/LSM) architecture against the legacy M3 (Ptrace) prototype and a Native Linux baseline.

  • Architecture: Sentinel M4 (Seccomp User Notification) vs. Legacy M3 (Ptrace).
  • Workload: High-frequency syscall stress test (1M+ ops).
  • Metric: Operations Per Second (OPS) and Relative Overhead.

Moving from Ptrace to Seccomp/LSM resulted in a 48x increase in throughput.

MetricNative LinuxSentinel M4 (Modern)Sentinel M3 (Legacy)
Throughput1,556,510 OPS1,366,558 OPS~28,000 OPS
Overhead0%~12%~5400%
Latency Cost0.13s2.31s>10.0s
xychart-beta
    title "Throughput Comparison (OPS - Higher is Better)"
    x-axis ["Sentinel M3 (Legacy)", "Sentinel M4 (Modern)", "Native Linux"]
    y-axis "Operations/Sec" 0 --> 1600000
    bar [28000, 1366558, 1556510]

The legacy M3 engine (Ptrace) paused the CPU for every syscall to switch contexts (Kernel User). This crushed throughput to just 28k OPS.

The modern engine filters events inside the kernel. Only “Critical” events trigger a notification. This allows Sentinel to retain ~88% of native throughput, making it viable for production workloads.

Verdict: The architectural shift to Kernel-Native enforcement (Seccomp/LSM) successfully bridged the gap between “Research Prototype” and “Production Engine.”