Odtnhj

I am writing a script for automatically computing average runtime.

First I need to run $ time ./foo.py for 100 times and save output to file time.txt (working)

$ for i in `seq 100`; do time ./foo.py; 2>> time.txt; done

Output looks as follows

time ./foo.py
real 0m0,030s
user 0m0,030s
sys 0m0,000s
[...]

Runtimes from different scripts are in the same file. Each entry starts with time ./foo.py, followed by 100 "triplets" of real, user and sys.

Now, if possible, I would love to have the script automatically compute the average runtime for each tested file by using all 100 "triplets", and neatly returning only one "mean triplet".

I have thought about maybe using awk to calculate the mean, like this

awk ' total += $2 END print total/NR ' time.txt

But the command would need to be adapted to fit my needs - after all, only the parts after the , (e.g. ,030s) may be used for computation and the s would also need to be disregarded.

Since I do not know how to achieve this objective, I thought to ask the community.

Any help is greatly appreciated.

It's easier if you tell time to output the time info in POSIX format:

awk '/^real/ totalReal += $2 /^user/ totalUser += $2 /^sys/ totalSys += $2 END print "realAvg " totalReal/(NR/4) "n" "userAvg " totalUser/(NR/4) "n" "sysAvg " totalSys/(NR/4) ' time.txt

Prints output as follows:

realAvg 12.62
userAvg 27
sysAvg 3.8

Explanation:

• Basically, tell awk to go through each line in the file, and if the line starts with real, add that to the totalReal variable, same for user and sys. So, basically, keep a running total of each of the three "types".


• At the end, simply print the the three running totals, divided by the number of lines divided by 4. This is because you want each "set" of 4 lines to count as 1 instance, and awk's NR just counts the number of lines.