|
|
|
|
|
Re: LTTng 2.0 [message #847701 is a reply to message #840195] |
Tue, 17 April 2012 15:13 |
Francois Chouinard Messages: 28 Registered: July 2009 |
Junior Member |
|
|
Hi,
After a bit of cleanup, the CTF code finally made it in the master branch and the nightly build (P2 repo: http://download.eclipse.org/technology/linuxtools/updates-nightly).
As mentioned before, only the basic views work for now (Events, Histogram, Statistics, Project). The others are a work-in-progress as we are fitting the kernel analysis views to our new state system management.
If you installed the LTTng 2.0 tracer on your target system (which can be your main system), you can create your own traces. Info for trace control and collection can be found here: http://wiki.eclipse.org/index.php/Linux_Tools_Project/LTTng/User_Guide#LTTng_Tracer_Control_.28LTTng_Tracer_Toolchain_v2.0.29.
If you don't have CTF traces handy or don't want to instrument your system, we could provide some samples on demand.
Before opening a trace you have to specify its type so the correct parser is used to process it, in this case the CTF parser. This can be done e.g. at import time by selecting "LTTng -> LTTng 2.0 Kernel Type" as the trace type. The trace icon should change to a penguin. Opening the trace (select and press "Enter" in the Project view) should populate the views immediately.
If you try to assign the legacy format parser ("LTTng -> Kernel Trace"), you will get an immediate error.
Hope this helps. For more info/clarifications, please consult the wiki/user guide (http://wiki.eclipse.org/index.php/Linux_Tools_Project/LTTng/User_Guide) or simply post a question.
Regards,
/fc
|
|
|
|
Re: LTTng 2.0 [message #850241 is a reply to message #849871] |
Thu, 19 April 2012 22:36 |
|
Hi,
I was looking at the trace, aand first off, thank you very much for the feedback. Your trace obviously contains more than two events.
I can see a few issues needing to be fixed and will get right to them.
expect a patch soon, if you are in a rush, I can send you the updates as they come.
|
|
|
|
Re: LTTng 2.0 [message #854320 is a reply to message #819337] |
Mon, 23 April 2012 21:59 |
|
I believe I have solved the issue, the patch should be in the tree in a day or so.
The issue was that I tested the system using traces of a 32 bit simple system. When a 64 bit trace is created, it has different alignments, and therefore, once your bitstream is no longer aligned, the parser gets lost and throws a metric boatload of exceptions.
I can send you a patch if you need it sooner.
Here is a shot of the working eclipse.
Thanks for the trace.
[Updated on: Mon, 23 April 2012 22:03] Report message to a moderator
|
|
|
|
Re: LTTng 2.0 [message #856314 is a reply to message #855142] |
Wed, 25 April 2012 14:58 |
|
With a state system, yes CPU usage can be done.
The reason: process 0 == swapper == idle. So whenever you're not in 0, you're running the CPU.
For memory it's more complicated. The reason: Shared memory, we can trace every malloc and free, and we can know the initial memory usage, but I would suggest using a UST tracer to poll your memory usage.
I think some people at Suse are working on a similar project. Check out the lttng-dev mailing list, they have more info.
|
|
|
Powered by
FUDForum. Page generated in 0.03791 seconds