creation, context switch time of process and thread

creation, context switch time of process and thread

[Srinivas Ramanan]

Can someone give the  information of following typical values in Kernel 3.x
in a 2Ghz machine with sufficient RAM

Process creation time ?
Process context switch time ?
pthread creation time ?
pthread context swich  time ?
kernel thread creation time ?
kernel thread context swich  time ?

What is the best way to measure these values?
I wrote  a simple test for measuring kernel thread context switch
A simple test like

1. get_time(start)
2. schedule();
3. get_time(end)
4. print( context sw time = (end-start) )

But it doesnt seem to work correctly.

Thanks.
Ramanan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20120524/153397a7/attachment.html 

creation, context switch time of process and thread

[刘智猷]

On 05/24/2012 08:25 PM, Srinivas Ramanan wrote:
> Can someone give the  information of following typical values in 
> Kernel 3.x
> in a 2Ghz machine with sufficient RAM
>
> Process creation time ?
> Process context switch time ?
> pthread creation time ?
> pthread context swich  time ?
> kernel thread creation time ?
> kernel thread context swich  time ?
>
> What is the best way to measure these values?
> I wrote  a simple test for measuring kernel thread context switch
> A simple test like
>
> 1. get_time(start)
> 2. schedule();
> 3. get_time(end)
> 4. print( context sw time = (end-start) )
>
> But it doesnt seem to work correctly.
>
> Thanks.
> Ramanan
>
>
>
>
>
> _______________________________________________
> Kernelnewbies mailing list
> Kernelnewbies at kernelnewbies.org
> http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies
Hi,

Check things supporting tracepoints like perf, systemtap and kprobe. 
They could help you a lot.

Regards,
LIU zhiyou
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20120524/65bd5353/attachment.html 

creation, context switch time of process and thread

[richard -rw- weinberger]

On Thu, May 24, 2012 at 2:25 PM, Srinivas Ramanan
<srinivas.ramanan@gmail.com> wrote:
> Can someone give the ?information of following typical values in Kernel 3.x
> in a 2Ghz machine with sufficient RAM
>
> Process creation time ?
> Process context switch time ?
> pthread creation time ?
> pthread context swich ?time ?
> kernel thread creation time ?
> kernel thread context swich ?time ?
>
> What is the best way to measure these values?

You can use events like sched_switch.
See: /sys/kernel/debug/tracing/events/sched/

-- 
Thanks,
//richard

creation, context switch time of process and thread

[Srinivas Ramanan]

On Fri, May 25, 2012 at 12:50 PM, richard -rw- weinberger <
richard.weinberger@gmail.com> wrote:

> On Thu, May 24, 2012 at 2:25 PM, Srinivas Ramanan
> <srinivas.ramanan@gmail.com> wrote:
> > Can someone give the  information of following typical values in Kernel
> 3.x
> > in a 2Ghz machine with sufficient RAM
> >
> > Process creation time ?
> > Process context switch time ?
> > pthread creation time ?
> > pthread context swich  time ?
> > kernel thread creation time ?
> > kernel thread context switch  time ?
>

Thanks for pointing out the tools. I will look in to them.
To confirm that i am measuring it right, can someone give some
(approximate) typical values for the above,
Assuming some 200 instructions in context switching  assembly code, i guess
the process context switch takes around 100ns. is this right?
I guess the process creation time will be greater than process context
switch time. correct?
Will the pthread context switch time be different from a kernel thread
context switch time?


thanks,
ramanan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20120525/4888518f/attachment.html 

creation, context switch time of process and thread

[Vladimir Murzin]

On Fri, May 25, 2012 at 10:59:16PM +0530, Srinivas Ramanan wrote:
> On Fri, May 25, 2012 at 12:50 PM, richard -rw- weinberger <
> richard.weinberger at gmail.com> wrote:
> 
> > On Thu, May 24, 2012 at 2:25 PM, Srinivas Ramanan
> > <srinivas.ramanan@gmail.com> wrote:
> > > Can someone give the  information of following typical values in Kernel
> > 3.x
> > > in a 2Ghz machine with sufficient RAM
> > >
> > > Process creation time ?
> > > Process context switch time ?
> > > pthread creation time ?
> > > pthread context swich  time ?
> > > kernel thread creation time ?
> > > kernel thread context switch  time ?
> >
> 
> Thanks for pointing out the tools. I will look in to them.
> To confirm that i am measuring it right, can someone give some
> (approximate) typical values for the above,
> Assuming some 200 instructions in context switching  assembly code, i guess
> the process context switch takes around 100ns. is this right?
> I guess the process creation time will be greater than process context
> switch time. correct?
> Will the pthread context switch time be different from a kernel thread
> context switch time?
> 
> 
> thanks,
> ramanan

> _______________________________________________
> Kernelnewbies mailing list
> Kernelnewbies at kernelnewbies.org
> http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies

Hi

Why don't you benchmark your environment with performance measuring
tools like LMBench[1]?

I guess that things are not easy... For instance, if process uses FPU
the state of FPU is usually saved on context switch, it could take
some time... It also possible that on some architectures it is
necessary to flush cashes and tlb on context switch,..
AFAIK processes and threads are handled by the Linux kernel in the
same way, i.e. there is no difference for scheduler between process
and thread.

[1]http://www.bitmover.com/lmbench/

Best wishes
Vladimir Murzin