From mboxrd@z Thu Jan  1 00:00:00 1970
From: mathieu.desnoyers@efficios.com (Mathieu Desnoyers)
Date: Mon, 7 Jan 2013 08:44:04 -0500
Subject: [lttng-dev] Cross-domain influence on overhead
In-Reply-To: <524C960C5DFC794E82BE548D825F05CF5BC8AEAD@EU-MBX-04.mgc.mentorg.com>
References: <524C960C5DFC794E82BE548D825F05CF5BC8AEAD@EU-MBX-04.mgc.mentorg.com>
Message-ID: <20130107134404.GA30165@Krystal>

* Oestman, Fredrik (Fredrik_Oestman at mentor.com) wrote:
> Hi,
> 
> I've noticed the following somewhat puzzling effect on both Intel and
> ARM architecture computers:
> 
> If all trace events are enabled in one domain only, either kernel or
> user space, and there are lots of static tracepoints instrumented in
> the other, then these inactive tracepoints start causing overhead of
> about the same magnitude as if they were active, as if trace events
> were enabled in this domain.
> 
> Is there any obvious explanation to this behaviour?

No, this would be really surprising.

Can you provide your test script, and more details about the way you
perform your benchmarks ? How is the overhead calculated ? From how many
tracepoints hit ? How many processes spawned/exec() done ? How many
tracepoints hit per process ?

Thanks,

Mathieu

> 
> Cheers,
> 
> Fredrik ?stman
> 
> http://go.mentor.com/sourceryanalyzer/
> 
> 
> _______________________________________________
> lttng-dev mailing list
> lttng-dev at lists.lttng.org
> http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev

-- 
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com