gpsprof — profile a GPS and gpsd, plotting latency information


gpsprof [-f plot_type] [-m threshold] [-n packetcount] [-t title] [-T terminal] [-d dumpfile] [-l logfile] [-r ] [-D debuglevel] [-h ] [[server[:port[:device]]]]


gpsprof performs accuracy, latency, and time drift profiling on a GPS. It emits to standard output a GNUPLOT program that draws one of several illustrative graphs. It can also be told to emit the raw profile data.

Information from the default spatial plot it provides can be useful for establishing an upper bound on latency, and thus on position accuracy of a GPS in motion.

gpsprof uses instrumentation built into gpsd.

To display the graph, use gnuplot(1). Thus, for example, to display the default spatial scatter plot, do this:

gpsprof | gnuplot -persist

To generate an image file:

gpsprof -T png | gnuplot >image.png


The -f option sets the plot type. The X axis is samples (either sentences with timestamps or PPS time drift messages). The Y axis is normally latency in seconds, except for the spatial plot. Currently the following plot types are defined:


Generate a scattergram of fixes and plot a probable-error circle. This data is only meaningful if the GPS is held stationary while gpsprof is running. This is the default.


Plot delta of system clock (NTP corrected time) against GPS time as reported in PPS messages.


Plot total latency without instrumentation. Useful mainly as a check that the instrumentation is not producing significant distortion. It only plots times for reports that contain fixes; staircase-like artifacts in the plot are created when elapsed time from reports without fixes is lumped in.


Plot instrumented profile. Plots various components of the total latency between the GPS's fix time fix and when the client receives the fix.

For purposes of the description, below, start-of-reporting-cycle (SORC) is when a device's reporting cycle begins. This time is detected by watching to see when data availability follows a long enough amount of quiet time that we can be sure we've seen the gap at the end of the sensor's previous report-transmission cycle. Detecting this gap requires a device running at 9600bps or faster.

Similarly, EORC is end-of-reporting-cycle; when the daemon has seen the last sentence it needs in the reporting cycle and ready to ship a fix to the client.

The components of the instrumented plot are as follows:

Fix latency

Delta between GPS time and SORC.

RS232 time

RS232 transmission time for data shipped during the cycle (computed from character volume and baud rate).

Analysis time

EORC, minus SORC, minus RS232 time. The amount of real time the daemon spent on computation rather than I/O.

Reception time

Shipping time from the daemon to when it was received by gpsprof.

Because of RS232 buffering effects, the profiler sometimes generates reports of ridiculously high latencies right at the beginning of a session. The -m option lets you set a latency threshold, in multiples of the cycle time, above which reports are discarded.

The -n option sets the number of packets to sample. The default is 100.

The -t option sets a text string to be included in the plot title.

The -T option generates a terminal type setting into the gnuplot code. Typical usage is "-T png" telling gnuplot to write a PNG file. Without this option gnuplot will call its X11 display code.

The -d option dumps the plot data, without attached gnuplot code, to a specified file for post-analysis.

The -l option dumps the raw JSON reports collected from the device to a specified file.

The -r option replots from a JSON logfile (such as -l produces) on standard input. Both -n and -l options are ignored when this one is selected.

The -h option makes gpsprof print a usage message and exit.

The -D sets debug level.

Sending SIGUSR1 to a running instance causes it to write a completion message to standard error and resume processing. The first number in the startup message is the process ID to signal.


gpsd(8), gps(1), libgps(3), libgpsd(3), gpsfake(1), gpsctl(1), gpscat(1), gnuplot(1).


Eric S. Raymond .