Question

I have a C++ application and i want to measure its performance on ARM board. The board is running ubuntu.

Currently i am considering valgrind and gprof to measure the performance.

What tools/techniques should i use to measure the performance?

Answer 1

Here are the biggies I ran across last time I had to do this:

• valgrind (only supported on cortex ARM processors.. boo)
• mprof (not so hot with threads?)
• gprof (not so hot with threads?)
• oprofile (requires kernel mods, but most modern kernels have it. I’ve used this under ARM.
• systemtap (recently ported to arm, looks awesome – like dtrace for Linux)
• strace and ltrace can actually be useful sometimes, although very high-level
• iostat et all as well if you want to kick it old school.
• Fair amount of information in /proc/ and /sys if you dig
• ioapps – IO tracing
• lsof is useful for tracking stuck sockets and file handles
• systat
• pmap
• iptraf
• tcpdump
• perftools – CPU and memory profiling
• bootchart
• QEMU can host ARM kernels / binaries, and can be instrumented from outside. It’s proven useful to me a couple times.

• Manual instrumentation using the gcc hooks

  void __cyg_profile_func_enter (void *, void *) __attribute__((no_instrument_function));
void __cyg_profile_func_exit (void *, void *) __attribute__((no_instrument_function));