How I Use Linux perf to Optimize the Performance of the Programe.

At beginning, the programe can only achieve 1.7Mbps under stress test after adding new features.

1.perf stat

$ sudo perf stat -p 27357
^C
 Performance counter stats for process id '27357':

       2120.354507      task-clock (msec)         #    0.013 CPUs utilized
            49,511      context-switches          #    0.023 M/sec
             2,721      cpu-migrations            #    0.001 M/sec
               381      page-faults               #    0.180 K/sec
     6,829,748,878      cycles                    #    3.221 GHz
     9,502,643,618      instructions              #    1.39  insn per cycle
     2,051,685,449      branches                  #  967.614 M/sec
        17,001,834      branch-misses             #    0.83% of all branches

     169.205295094 seconds time elapsed

It indicates that this doesn’t seem to be a CPU-bound problem as the value of task-clock (msec) for CPU-bound one can achive almost 1[1].

2.perf top

$ sudo perf top -p 27357

3.perf record && perf report

sudo perf record -g -F 50000 -e cpu-clock -p 27357

A file named perf.data is generated.
sudo perf report -i perf.data

4.References

1.https://www.ibm.com/developerworks/cn/linux/l-cn-perf1/index.html
2. http://www.blogjava.net/qileilove/archive/2013/09/04/403646.html
3. https://www.cnblogs.com/happyliu/p/6142929.html