audria is a little tool designed to read various program statistics from the /proc filesystem, including:
- current and average CPU usage
- virtual memory usage (peak, resident set, swap etc.)
- IO load (current/total read/writes with and without buffers)
- time spent in user/system mode
- page faults
- and other information
audria can be used to monitor a single process, a group of processes or all currently running processes. Furthermore it is possible to specify a command to execute in order to watch a process directly from startup. In contrast to some other tools, audria is designed to run in batch mode at very short intervals. This allows very detailed inspections of the resource usage of a process.
audria has several command line arguments:
PID(s) PID(s) to monitor
-a monitor all processes
-d delay delay in seconds between intervals (default: 0.5), specify '-1' to use
2 * kernel clock tick rate (2 * 1/100 on most systems), note: values equal or below
the kernel clock tick rate will lead to bogus values for the 'CurCPUPerc' field
-e cmd program to execute and watch, all remaining arguments will be forwarded
-f fields names of fields to show, separated by comma (default: all)
-k show kernel threads (default: false)
-n num number of iterations before quitting (default: unlimited)
-o file file to write output to instead of stdout, will append to existing files,
if file is '-' then output will be written to stdout (default)
-r acquire real-time priority (lowest niceness, highest scheduling priority),
usually requires root privileges or the CAP_SYS_NICE capability
-s include self in list of processes to monitor
-h print this help and exit
Usually it is sufficient to just pass the PID you want to monitor and optionally the name of the output file:
audria $(pidof myProgram) -o data.txt
For deeper inspections you might want to watch the program directly from startup and at a higher interval:
audria -d -1 -o data.txt -e myProgram -myProgramArgument
You can also specify which fields to show:
audria -f Name,CurCPUPerc,Threads,VmSizekB,CurReadBytesPerSec,CurWrittenBytesPerSec -a
audria generates a CSV-like output which is suitable for plotting.
For the gnuplot software you can use the example file example.plot as a start.
It will plot several graphs based on the generated data, assuming this data has been stored as data.txt.
Just install gnuplot and run gnuplot example.plot to generate the graphs.
These are some example plots created with gnuplot from audria's output. A firefox process was monitored right from the start:
The first plot shows the CPU (current + average) and memory (allocated + actually used) usage.
Seems like IO is the bottleneck at startup as the CPU is only occasionally active:
The second plot visuializes the IO load, especially the current number of read/written bytes/s and read/write calls:
In the last plot we can see the relative time the process spend in user and kernel mode, as well as the current number of threads:
