Documentation

PingZen monitors report latency for PING and HTTP/HTTPS checks. Below is exactly how we measure, what each metric means, and how our numbers compare to industry-standard tools (Linux ping, curl, icmplib, fping).

Zoom

PING (ICMP) — Kernel-Accurate RTT

For Ping monitors we shell out to the system ping binary (same as Datadog and most probes) and parse its statistics line. All reported RTT values come directly from the kernel — not from any Python timestamp — so they are identical to what you would see running ping on the probe host yourself.

How we measure (step by step)

1. Run ping -c N -i 0.2 <host> as an asyncio subprocess (N = configurable packet count, default 4).
2. iputils ping uses gettimeofday() for send-side timestamp + kernel SO_TIMESTAMP control message (cmsg) for receive-side — see iputils source.
3. Parse the rtt min/avg/max/mdev line from stdout (Linux) or round-trip min/avg/max/stddev (macOS).
4. Store min_rtt, avg_rtt, max_rtt, jitter (mdev), and packet_loss as floats in the check_results table.
5. The integer response_time_ms column is set to round(avg_rtt) with a 1 ms floor — sub-ms RTTs round to 1 ms instead of 0, which reads as “no response”.
6. Subprocess fork+exec overhead (~10 ms wall-clock) is discarded — it never enters the RTT figures.

Stored fields

HTTP / HTTPS — Time to First Byte with Breakdown

For HTTP/HTTPS monitors we use httpx.AsyncClient with the trace extension hook to capture individual phase timestamps. We record wall-clock total plus four sub-event durations. Convention matches curl time_total and time_starttransfer.

Per-check timing fields

Industry Standards We Follow

PingZen measurements are RFC-compliant and align with major tools:

• RFC 2681 — Round-trip Delay Metric for IPPM: our ICMP path via system ping qualifies as a Type-P-Round-trip-Delay measurement.
• Datadog Synthetic ICMP tests use ping subprocess + stdout parse — identical to PingZen.
• Prometheus blackbox_exporter uses Go time.Now() userspace timestamps, no SO_TIMESTAMPING — PingZen has the same precision class via kernel recv-side cmsg.
• SmokePing / fping: batched C raw sockets for efficiency. Same metric semantics (min/avg/max/jitter/loss), different implementation.
• RIPE Atlas: ICMP echo with internal firmware timestamping. Precision not publicly documented.
• icmplib (Python): pure-Python raw socket. Validated against PingZen subprocess — RTT matches within 0.1 ms.
• SO_TIMESTAMPING kernel hardware timestamps exist for nanosecond precision but are overkill for application-level uptime monitoring (and unsupported in containerized environments).

Common Questions