Hi Sébastien,
You could always write custom scripts using grep and sed, but those tend
to be a bit inflexible. ;)
Have you heard about Trace Compass [1] (previously known as TMF, or "The
Eclipse LTTng plugin") ? It's a generic trace viewer and analyzer based
on Eclipse, and it supports LTTng traces. It's not officially "out" yet,
because the project itself is still being set up. We don't have any
download links on the new website, but in the meantime there are
relatively recent builds at [2].
There are some default basic views for UST traces, like event list,
statistics, memory usage if you have the right events enabled, etc. But
in your case it seems you need something more specific. Depending on
your level of wanting to dig into it, you could write an "analysis
module" that would receive each event in the trace, and then check if
they're of type "actor_receive" or "actor_send", and calculate the delay
between two consecutive ones. Then aggregate the results and possibly
print the worst offenders in a view (or even to the console if it's
sufficient).
See [3] for how to setup the dev environment, and [4] for the analysis
framework documentation.
Let us know if you need more information!
Cheers,
Alexandre
[1] http://eclipse.org/tracecompass
[2] http://secretaire.dorsal.polymtl.ca/~gbastien/TracingRCP/TraceCompass/
[3] https://wiki.eclipse.org/Trace_Compass/Development_Environment_Setup
[4]
http://wiki.eclipse.org/Linux_Tools_Project/TMF/User_Guide#Analysis_Framework
On 10/17/2014 06:23 PM, Boisvert, Sebastien wrote:
Bonjour,
First, thank you for LTTng-UST. This is very useful and convenient.
I just got started today using LTTng (LTTng-UST) for tracing a HPC application
that I am working on (I am a postdoc). I am impressed by how easy LTTng is to
use it.
In my system, an actor message is represented by a pair
<message_actor_source, message_number>.
I want to list all messages that have a high delivery time
(message:actor_receive - message:actor_send).
I am doing this to get the messages of one actor (actor 1000019):
[boisvert@bigmem biosal]$ babeltrace ~/lttng-traces/auto-20141017-181240|grep
"message_source_actor = 1000019" > actor_1000019
Then, I can look at one message with (message <1000019, 14>):
[boisvert@bigmem biosal]$ grep "message_number = 14," actor_1000019
[18:12:43.647017211] (+0.000005110) bigmem.knoxville.kbase.us
message:actor_send: { cpu_id = 30 }, { message_number = 14, message_action =
31592, message_count = 8, message_source_actor = 1000019,
message_destination_actor = 1000059, message_source_node = -1,
message_destination_node = -1 }
[18:12:43.647025249] (+0.000002860) bigmem.knoxville.kbase.us
message:actor_receive: { cpu_id = 49 }, { message_number = 14, message_action =
31592, message_count = 8, message_source_actor = 1000019,
message_destination_actor = 1000059, message_source_node = 3,
message_destination_node = 3 }
If I substract the times:
irb(main):003:0> (43.647025249-43.647017211)*10**9
=> 8038.00000426236
This message (<1000019, 14>) required 8038 ns for the delivery. This one is
fine.
So basically my question is:
Is there an easy way to analyze these tracepoint files ?
_______________________________________________
lttng-dev mailing list
[email protected]
http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
_______________________________________________
lttng-dev mailing list
[email protected]
http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
