TIME command

When running a program within the time command three values are given: REAL, USER and SYSTEM. For benchmarking purposes, which value do I use if I want execution time irrespective of machine overhead?
rsorrentAsked:
Who is Participating?

Improve company productivity with a Business Account.Sign Up

x
 
elfieConnect With a Mentor Commented:
Real time is the differnce between the starting time and ending time.(elapsed time)

User time is the cpu time spend in 'user' coding.
system time is the time spend in executing 'system' code.

If you run on a clean system (no overhead) you will see that adding user and system time will give you (approximately) the real time.

If you have no 'clean' system, just take user and system time, and it will give you a good indication for benchmarking.

On a heavy loaded system, 'system' time can increase due to a higher number of page faults, swap out/in, and other interruptions.

Always keep in mind that there is always a difference between a first run and the next executions of a program. Executing a program a second time, mostly means thah the program is still i memory, and thus requires less startup time.
0
 
JYoungmanCommented:
Any of the three, still, depending on what you mean by "machine overhead".  Perhaps if I explain the three times, you'll be able to choose the one you want.

When a program runs, it uses the CPU for up to 100% of the time it is running.  If it spends most of its time waiting for data from disks or terminals, or waiting for disk writes to complete, it may only use 1% of the CPU's time.  In this case, REAL time may be one second (the program took 1s to finish) but (USER+SYSTEM) may only be 0.01s.

USER time is time your program spends actually using the CPU to execute its own instructions.  SYSTEM time is time spent by the kernel using the CPU on behalf of that program, to fulfil requests that the program makes.   Here is an example program:-

[james@noisy james]$ cat us.c
#include <sys/time.h>
#include <unistd.h>

int main()
  {
  int i;
  struct timeval tv;
  struct  timezone tz;
  for (i=0; i<1000000; ++i)
    gettimeofday(&tv, &tz);
  return 0;
  }
[james@noisy james]$ make CFLAGS="-O2" LDLIBS= us
cc -O2  -g  us.c   -o us
[james@noisy james]$ time ./us
0.60user 1.80system 0:02.41elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (60major+7minor)pagefaults 0swaps

Here we see that most of the execution time was spent using the CPU, but 1.8s of that (75%) was SYSTEM (i.e. kernel) time.

If you're benchmarking a database, wall-clock time taken to complete a task may be the best measurement.   On the other hand, for a compute-intensive task, USER or (USER+SYSTEM) may be the best measures.


0
Question has a verified solution.

Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.

Have a better answer? Share it in a comment.

All Courses

From novice to tech pro — start learning today.