This is the mail archive of the
systemtap@sourceware.org
mailing list for the systemtap project.
Re: "lttng comparison with other tracers" page
* Frank Ch. Eigler (fche@redhat.com) wrote:
> Hi -
>
> On Wed, Mar 18, 2009 at 11:14:55AM -0400, Mathieu Desnoyers wrote:
>
> > [...] Have you measured the performance impact of disabling
> > interrupts around the probe ?
>
> Yes. Among others, we received reports where oprofile was indicating
> more traffic at interrupt-enable time along with somewhat lower
> interrupt-handling-latency-related performance.
>
Was oprofile triggered by NMIs or by standard interrupts ? If it is the
second option, then it's very normal to have lots of events at
interrupt-enable time and no events at all distributed within the
interrupt disabled section.
Are you sure that the performance regression you got was caused by added
interrupt latency ? How much time can be typically spent in a systemtap
probe (order of magnitude) ?
> > Basically, it sounds like you are shipping a car without brakes nor
> > air bags simply because it makes the car too heavy and therefore too
> > slow. I don't think trading stability and robustness for speed is a
> > good tradeoff.
>
> We are doing neither - stability and robustness are not at risk by
> this option. You noticed that the system responded with a clean
> script stop.
>
I guess we have totally different definitions of "clean" then. Being
unable to report correctly events coming from reentrant execution
contexts is at the very least misleading to the user.
> We're starting to see this happen more now only because the kprobes
> machinery blocked reentrancy in other ways for us, so we never had to
> think hard about what is a good value for the MAXSKIPPED parameter.
I am still not convinced that irqoff is not a good solution. Skipping
events seems like the wrong course of action to me.
Mathieu
>
> - FChE
>
--
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68