Re: Oprofile JIT and JBoss

[soila <sp...@ec...>]

Thanks Maynard,

I do not think the servlet code gets JITed because the
JVMPI_EVENT_COMPILED_METHOD_LOAD is not being called even though the
JVMPI_EVENT_CLASS_LOAD is captured for the servlet classes.

Do you have any pointers on how to implement the call-graph functionality
for JIT support?

Soila

Maynard Johnson wrote:
> 
> soila wrote:
>> Hi,
>>
>> I tried using the JVMPI tool available in oprofile 0.9.4 to profile an
>> EJB
>> application running on JBoss 3.2.8 and Java 1.4.2 (I am not using the
>> latest
>> version of JBoss and Java for legacy reasons). 
>>
>> The output of opreport does not capture application-level function calls
>> made by the servlets, but rather outputs only JVM functions. The oprofile
>>   
> Does the code executed by the servlets get JITed?  Only JITed code is
> supported by OProfile.
>> call-graphs for the Java functions are also not detailed and only include
>> a
>> reference to self, e.g.,
>>
>> -------------------------------------------------------------------------------
>> 434       0.1133  32747.jo                 java                     int
>> java.util.HashMap.hash(java.lang.Object)
>>   434      100.000  32747.jo                 java                     int
>> java.util.HashMap.hash(java.lang.Object) [self]
>> -------------------------------------------------------------------------------
>>   
> Chapter 4, Section 2.3.2 of the OProfile user manual (doc/oprofile.html
> in the RC1 tar ball) describes the limitation of callgraph output in
> regards to the new JIT support.  Removing this limitation is in the TODO
> list, but it will definitely be a non-trivial effort.  If you could work
> on providing such a patch, that would be great.
>> I run oprofile with --separate=thread. Are there additional events I
>> should
>> listen for in the JVMPI profiler, e.g., JVMPI_EVENT_METHOD_ENTRY2 and
>> JVMPI_EVENT_METHOD_EXIT?
>>   
> Besides the likelihood that this would probably impose a lot of extra
> overhead (and possibly affect overall system performance), simply adding
> these events to the JVMPI agent wouldn't help.  The agent and the JIT
> support framework are responsible for creating an ELF file that
> represents the JITed code.  The event method entry/exit information
> would have to be stored in some other file, and then the callgraph
> processing code of OProfile would have to be "taught" how to make use of
> the information.  Even with that, you would still need to correlate the
> right call stack with a given sample.
> 
> Regards,
> -Maynard
>> Thanks,
>>
>> Soila
>>   
> 
> 
> 
> -------------------------------------------------------------------------
> Check out the new SourceForge.net Marketplace.
> It's the best place to buy or sell services for
> just about anything Open Source.
> http://sourceforge.net/services/buy/index.php
> _______________________________________________
> oprofile-list mailing list
> opr...@li...
> https://lists.sourceforge.net/lists/listinfo/oprofile-list
> 
> 

-- 
View this message in context: http://www.nabble.com/Oprofile-JIT-and-JBoss-tp18032094p18188286.html
Sent from the oprofile-list mailing list archive at Nabble.com.