Re: [PATCH] perf record: mmap output file - RFC

[Jiri Olsa <jolsa@redhat.com>]

On Sun, Sep 22, 2013 at 08:05:59PM -0600, David Ahern wrote:
> When recording raw_syscalls for the entire system, e.g.,
>     perf record -e raw_syscalls:*,sched:sched_switch -a -- sleep 1
> 
> you end up with a negative feedback loop as perf itself calls
> write() fairly often. This patch mmap's the file in chunks of 64M
> at a time and copies events from the event buffers to the file
> avoiding write system calls.

moved processing into userspace:

    17.24%  -17.10%  libpthread-2.15.so  [.] __write_nocancel                    
     ...
     0.07%   +0.64%  libc-2.15.so        [.] __memcpy_sse2                               
     0.02%  +51.84%  libc-2.15.so        [.] __memcpy_ssse3_back                         
     0.01%   +0.34%  libc-2.15.so        [.] __mempcpy_sse2                              
     ...
> 
> Before (with write syscall):
> 
> perf record -o /tmp/perf.data -e raw_syscalls:*,sched:sched_switch -a -- sleep 1
> [ perf record: Woken up 0 times to write data ]
> [ perf record: Captured and wrote 81.843 MB /tmp/perf.data (~3575786 samples) ]
> 
> After (using mmap):
> 
> perf record -o /tmp/perf.data -e raw_syscalls:*,sched:sched_switch -a -- sleep 1
> [ perf record: Woken up 31 times to write data ]

                          ^^^^^^^^
but it's still faster, since we finally get perf a chance to sleep ;-)

new time:
	real    0m30.392s
	user    0m0.041s
	sys     0m0.389s

old time:
	real    0m32.235s
	user    0m3.080s
	sys     0m14.444s


> [ perf record: Captured and wrote 8.203 MB /tmp/perf.data (~358388 samples) ]
> 
> Before I get too far down this path I wanted to get comments on the approach.

I think it's worthwhile doing this

jirka