Empirical profiling runs
Sub-millisecond runtime tracing under 100k req/s load.
12 µs
P99 Latency Overhead
32 MB
Resident Memory Footprint
0
Heap GC Pauses
Resource consumption
0.14%
CPU Share (100k RPS)
32 MB
Resident Memory
1.2 MB/s
Disk IO Throughput
0 drops
Ring Buffer Overrun
Test bed specification
Hardware environment and kernel flags
Benchmarks execute on isolated AWS c6i.16xlarge instances running Linux kernel 6.2 with eBPF ring buffer aggregation enabled. Synthetic load is generated via wrk2 across 128 concurrent TCP connections against a mock microservices mesh.
Memory profiles are captured using jemalloc heap profiling over 48-hour soak tests to verify zero memory creep or uncollected pointer references under heavy trace volume.
Validate telemetry overhead on your staging cluster.
Pull the open-source benchmark runner, compile the collector binary, and run deterministic verification tests locally.