Re: Perf trace event parse errors for KVM events
From: Steven Rostedt
Date: Thu Jun 03 2010 - 17:57:16 EST
On Tue, 2010-06-01 at 15:39 +0300, Avi Kivity wrote:
> On 06/01/2010 02:59 PM, Steven Rostedt wrote:
> I meant that viewing would be slowed down. It's an important part of
> using ftrace!
>
> How long does the Python formatter take to process 100k or 1M events?
>
I finally got around to testing this.
I ran a trace on lock_acquire, and traced 1,253,296 events.
I then created a python plugin to analyze the trace:
----
def lock_acquire(trace_seq, event):
t = ''
r = ''
if int(event['flags']) & 1:
t = 'try'
if int(event['flags']) & 2:
r = 'read'
trace_seq.puts('t %x %s%s%s' % (
event['lockdep_addr'], t, r,
event['name']))
def register(pevent):
pevent.register_event_handler("lock", "lock_acquire", lock_acquire)
----
Disclaimer, I'm not a python expert, and I'm sure the above python code
sucks.
[root@ixf9 trace-cmd.git]# time ./trace-cmd report -N >/dev/null 2>&1
real 0m4.653s
user 0m4.234s
sys 0m0.419s
* -N keeps trace-cmd from loading any plugins.
[root@ixf9 trace-cmd.git]# time PYTHONPATH=`pwd` ./trace-cmd report >/dev/null 2>&1
real 0m53.916s
user 0m53.047s
sys 0m0.859s
Yes, running a python interpreter is a bit more expensive. It took 4
seconds to read the million events with plain C, but 53 seconds to read
it in python.
That said... This would only affect you if you were writing this to a
file. I doubt that you would notice this if you were scanning the trace
with less.
Also, I kicked this off in kernelshark, and it made no difference that I
can see. This is because kernelshark only evaluates the viewable area of
the screen.
-- Steve
--
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/