I keep asking myself the same question. Using ping_exporter (which is basically just a wrapper around go-ping, pinging between two systems on my network, I see ~230 µs for IPv6 and ~210 µs for IPv4.

Running ping on the command line between the same systems gives a mean latency of 65 µs for IPv4 and 68 µs for IPv6.

At least part of this is ping using SIOCGSTAMP by default; this reports the kernel's packet receive time instead of calling gettimeofday during the packet processing process. You can disable this with ping via -U; on my system this adds an extra 70 µs or so to the RTT displayed. Since that's using fairly simple C, I could absolutely see calculating timestamps in Go adding an extra 100 µs.

There are two ways to get this data from the kernel; either SIOCGSTAMP (via ioctl) or SO_TIMESTAMP (via recvmsg). It looks like they're both possible in Go, but neither is terribly portable or clean. See go-pinger's source for SIOCGSTAMP and pcapgo's source for SO_TIMESTAMP.

Why is the return time longer than the native PING? #20

Description

Activity

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions