Re: Is lttng able to trace a process's user time and kernel time ?

* yuanbor ([email protected]) wrote:
> hello everybody,
> Now, I am working on operating system which is designed to tackle multicore problem. There is a powerful platform in my lab. The platform has 2 processors, each has 4 cores. The total physical memory is 8GB. I also have some benchmarks, such as, SPECcpu2006, SPECweb2005, SPECjbb2005, SPECmpi2007 and TPCH on oracle 11g.
> What I want first is the ratio of user time to kernel time of these benchmarks. I used lttng before, but find that it add some excess cost to the system, so it may decrease the reliability of the result.
> 
> In the end, my question is that can I find a more lighter weight matter by lttng to get my point?
> 

Since we trace kernel entry/exit at syscall, irq, etc, _within_ the
kernel code, we will account more userspace time than what a profiler
could see.

You also have to consider that the profiling approaches are _not_ exact
and use sampling techniques. It may have problems taking correctly into
account periodic events.

Mathieu

> Any reply would be nice.
> Thank you very much.

> _______________________________________________
> Ltt-dev mailing list
> [email protected]
> http://listserv.shafik.org/mailman/listinfo/ltt-dev


-- 
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68