* Steven Rostedt <rostedt@goodmis.org> wrote:

quoted text
> On Wed, 2010-05-12 at 18:59 +0200, Ingo Molnar wrote:
> > * Steven Rostedt <rostedt@goodmis.org> wrote:
> > 
> > > On Wed, 2010-05-12 at 16:48 +0200, Frederic Weisbecker wrote:
> > > > On Wed, May 12, 2010 at 03:37:27PM +0200, Pierre Tardy wrote:
> > > 
> > > > But we don't yet support trace_printk in perf. May be we could wrap
> > > > them in trace events.
> > > 
> > > Hmm, do we really want to do that?
> > > 
> > > We really need to get the perf and ftrace trace buffers combined. I 
> > > understand why perf chose to do the mmap buffers for the counting, but 
> > > for live streaming, it is very inefficient compared to splice.
> > 
> > The thing is that for a very long time ftrace didnt have splice support 
> > and survived just fine. Even today most of the ftrace usage isnt utilizing 
> > splice.
> 
> Actually, trace-cmd implements the splice interface and is used by several 
> people. I find myself using trace-cmd 90% of the time that I use ftrace, 
> specifically because of this speedup.

i know, but most people still use /debug/tracing/ bits not trace-cmd.

quoted text
> > Yes, splice might help in some situations but on average it's an 
> > independent speedup on the order of magnitude of a few percents, not a 
> > 'must have' item.
> 
> I'll have start running benchmarks to see what the actual speed up is. I'm 
> guessing it may be more than a few percent. It allows for zero copy overhead 
> and reuse of the data page.

Make sure you measure it in the context of a full app like PyTimechart.

You can measure the overhead using perf stat ;-)

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