====== Usage Profiling and benchmarking ======
===== Shell =====

''/usr/bin/time'' ((**NB.** not bash built-in ''time'', need explicit path)) can report total run time ("clock on the wall") and memory (''RSS'' = Resident Set Size) 
<code>
/usr/bin/time -v perl -e '"a"x(1024**3);' |& grep -Pi 'max|wall'
#        Elapsed (wall clock) time (h:mm:ss or m:ss): 0:00.40
#        Maximum resident set size (kbytes): 1053908
</code>

==== Super Computer Slurm ====

When run with [[:tools:slurm]], eg. [[:tools:psc]]
<code>
 sacct -j $jobid --format=JobID,MaxRSS,AveRSS,TotalCPU,Elapsed
</code>

==== hyperfine ====

For smaller jobs, ''hyperfine'' runs many time for a more rigorous estimate.
<code>
hyperfine 'sleep .5' 'sleep .55'
</code>
<code>
Benchmark 1: sleep .5
  Time (mean ± σ):     502.3 ms ±   0.6 ms    [User: 1.6 ms, System: 0.8 ms]
  Range (min … max):   501.7 ms … 503.2 ms    10 runs
 
Benchmark 2: sleep .55
  Time (mean ± σ):     552.6 ms ±   0.5 ms    [User: 1.5 ms, System: 1.0 ms]
  Range (min … max):   551.8 ms … 553.4 ms    10 runs
 
Summary
  sleep .5 ran
    1.10 ± 0.00 times faster than sleep .55
</code>

===== R =====
[[:tools:R]] can use ''microbenchmark''
<code>
microbenchmark::microbenchmark(sqrt100 = sqrt(runif(100)))

#Unit: microseconds
#    expr   min    lq    mean median    uq    max neval
# sqrt100 5.769 5.911 6.82513  6.071 6.445 46.451   100
</code>

===== python =====
<code python>
import timeit
timeit.timeit('numpy.sqrt(numpy.random.uniform(size=100))',
              setup='import numpy')
# 3.9988546459935606
</code>