[ltt-dev] performance results I measured with the latest lttng patches

[Mathieu Desnoyers]

* Michael Rubin (mrubin at google.com) wrote:
> On Wed, Sep 24, 2008 at 8:14 AM, Mathieu Desnoyers
> <compudj at krystal.dyndns.org> wrote:
> > yeah, about, say.. unnoticeable ? :)
> 
> This is great news. I was not sure how closely aligned we are.
> 
> What are the goals for when tracing is enabled? At that point
> unnoticeable seems unrealistic. Is there a perf impact where you say
> good enough?
> 

I would say I tend to expect something around the lines of 5%
performance impact on a flight-recorder mode trace. The current LTTng
might suffer from TLB entries trashing because it uses vmap'd memory,
something that should be looked at.

> > I have a dual-quad cores at my lab, I'll run some tbench tests on this
> > setup to try to figure out what went wrong. I guess there might be some
> > option enabled when it shouldn't, or maybe I made some sort of stupid
> > error.... :) Anyway, that should not be too hard to find, hopefully.
> >
> 
> Great. Thanks a ton. There are a lot of folks at Google hungry for this feature.
> 
> The more improvements in performance the easier it is for me to throw
> resources and hopefully contribute to your efforts too.
> 

I have to look into some tracepoint detail.. it seems that the kernel
code in handle_irq_event is very sensitive to instrumentation.

Mathieu

> Thanks again,
> 
> mrubin
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68