Sean McIntyre · designing technology to create art

Standard hardware timing inaccuracies

Don't bother trying to find the difference of two events occurring on standard computer hardware if you are looking for sub-microsecond precision.

I learned this today after putting two of my WNDR3700v2 routers running OpenWRT into monitor mode to look at the packets floating in the air around me. My goal was to triangulate the origin of the packets.

I managed to measure the arrival time of a packet on the router down to the microsecond. Unfortunately, the processors report times accurate up to about 1 ms. Light can travel 300 km in 1 ms, so there's no real chance that I'll be able to pinpoint where a signal originates using this method!

I'm not the first person to realize that timing is inaccurate without dedicated hardware keeping track of time. See this thread.

Jitter plot

The above graph plots the delta in microseconds (Y-axis) between many packet receive times (X-axis) on two separate routers. The bulk of the packets are between -7000 and -8000, with a few outliers. I used libpcap and gettimeofday() to measure this, comparing hashes of packet payloads to check if the packets are the same.

comments powered by Disqus