Krdytkyu

I hope this is the correct place to ask this question.

I would like to benchmark a program using some test instances. I’m primarily interested in the cpu time the program took to execute and to this end, I’ve found the time command that gives me the real, user and sys time. As I’m quite new to linux, I’m not sure which of these to take. For me, user and sys seem to be my best bet, but I’m not quite sure which one to take. Does it make sense to add them up to get the total cpu-seconds or is user the most meaningful in my application?

Any help appreciated!

By default, there are two time commands on your system:

• a bash shell keyword
  
  This is called when you run time, you can get help about it with help time. If you want to tweak its output, read about the TIMEFORMAT variable in man bash.


• a binary /usr/bin/time
  
  If you want this you need to call it with its full path, you can get help about it with man time.

The /usr/bin/time binary comes with a nice -f, --format option which lets you choose the information in its output. The mentioned manpage has the full list of specifiers under FORMATTING THE OUTPUT, here are just some suggestions that may be relevant for you:

K Average total (data+stack+text) memory use of the process, in Kilobytes.
S Total number of CPU-seconds used by the system on behalf of the process (in kernel mode), in seconds.
U Total number of CPU-seconds that the process used directly (in user mode), in seconds.

If S and U is what’s important in your case and you want those two on separate lines, your command line could look like this:

/usr/bin/time -f "%Sn%U" /path/to/program

As S and U are available for bash’s time as well, this translates to:

TIMEFORMAT=$'%Sn%U' && time /path/to/program

It’s worth noting that the bash keyword is able to output milliseconds while the binary does only output centiseconds. The keyword may also be faster, as no external program needs to be called.

For an overview on the timing topic, I recommend reading man 7 time.