Re: [PATCH 5/5] perf report: Support callchains with relativeoverhead rate

[Frederic Weisbecker]

On Sun, Jul 05, 2009 at 10:34:00AM +0200, Ingo Molnar wrote:
> 
> * Frederic Weisbecker <fweisbec@xxxxxxxxx> wrote:
> 
> > The current callchain displays the overhead rates as absolute:
> > relative to the total overhead.
> > 
> > This patch provides relative overhead percentage, in which each
> > branch of the callchain tree is a independant instrumentated object.
> > 
> > You can produce such output by using the "relative" mode
> > that you can lower in r, re, rel, etc...
> > 
> > ./perf report -s sym -c relative
> > 
> > Example:
> > 
> >      8.46%  [k] copy_user_generic_string
> >                 |
> >                 |--52.01%-- generic_file_aio_read
> >                 |          do_sync_read
> >                 |          vfs_read
> >                 |          |
> >                 |          |--97.20%-- sys_pread64
> >                 |          |          system_call_fastpath
> >                 |          |          pread64
> >                 |          |
> >                 |           --2.81%-- sys_read
> >                 |                     system_call_fastpath
> >                 |                     __read
> >                 |
> >                 |--39.85%-- generic_file_buffered_write
> >                 |          __generic_file_aio_write_nolock
> >                 |          generic_file_aio_write
> >                 |          do_sync_write
> >                 |          reiserfs_file_write
> >                 |          vfs_write
> >                 |          |
> >                 |          |--97.05%-- sys_pwrite64
> >                 |          |          system_call_fastpath
> >                 |          |          __pwrite64
> >                 |          |
> >                 |           --2.95%-- sys_write
> >                 |                     system_call_fastpath
> >                 |                     __write_nocancel
> > [...]
> 
> Wow, this is extremely intuitive and powerful looking!
> 
> It's basically a fractal structure: each sub-graph looks like a 
> full-blown profile in itself. Thus the overhead of individual 
> components of the graph profile can be analyzed without having to 
> think in small numbers.
> 
> The above example shows it particularly well - it shows that in 
> regard to generic_file_buffered_write() overhead, the system is 
> doing 97% sys_pwrite64() calls and 3% sys_write() calls.
> 
> Thus i took the liberty to change your last patch in two ways: i 
> renamed 'relative' to 'fractal' (it was not a proper counterpart to 
> 'graph' anyway - we have no 'absolute' output mode name either), and 
> i changed it to be the default output mode. This stuff rocks!



Ok.

I first planned to add a submode, or more likely a parameter structured
like the following:

perf report -c layout,min,mode

where layout=flat|graph
and   mode=abs|rel

But relative flat doesn't make sense.
fractal is nice, it's just a pity that "graph" mode doesn't tell much
about it's absolute measure context.
 
> absolute-graph and flat mode can be displayed too, via the option, 
> as usual.
> 
> 	Ingo

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/