[ltt-dev] Libtraceread sequential reading of events with the same timstamp

[ltt-dev] Libtraceread sequential reading of events with the same timstamp

[Matthew Khouzam]

Hello world,
When you have different events in different tracefiles of the same trace 
that have the same timestamp, is the order of reading supposed to be 
pre-determined?

An example to better illustrate my question.

If at 100ns you have a kernel softIrqEntry and a fs open that occur. 
Would you have a prefered order of reading the trace, and could this 
affect the state machine.

thanks,

Matthew

[ltt-dev] Libtraceread sequential reading of events with the same timstamp

[Mathieu Desnoyers]

* Matthew Khouzam (matthew.khouzam at ericsson.com) wrote:
> Hello world,
> When you have different events in different tracefiles of the same trace  
> that have the same timestamp, is the order of reading supposed to be  
> pre-determined?
>
> An example to better illustrate my question.
>
> If at 100ns you have a kernel softIrqEntry and a fs open that occur.  
> Would you have a prefered order of reading the trace, and could this  
> affect the state machine.

If two events happen in different trace files at exactly the same time,
there is nothing we can tell about the ordering. So their relative order
is not important, even though it could affect the state machine in rare
cases. Ideally, the state machine should become clever enough to detect
these cases where relative ordering of events is unknown based on
timestamps delta being under a certain threshold, or based on other
instrumentat (interrupt execution surrounding the event). I'm not saying
that LTTV and TMF are doing a particularly good at this at the moment --
it should be considered as something to improve.

Thanks,

Mathieu

-- 
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com

[ltt-dev] Libtraceread sequential reading of events with the same timstamp

[Alexandre Montplaisir]

On 10-11-30 03:30 PM, Mathieu Desnoyers wrote:
> * Matthew Khouzam (matthew.khouzam at ericsson.com) wrote:
>> Hello world,
>> When you have different events in different tracefiles of the same trace
>> that have the same timestamp, is the order of reading supposed to be
>> pre-determined?
>>
>> An example to better illustrate my question.
>>
>> If at 100ns you have a kernel softIrqEntry and a fs open that occur.
>> Would you have a prefered order of reading the trace, and could this
>> affect the state machine.
> If two events happen in different trace files at exactly the same time,
> there is nothing we can tell about the ordering. So their relative order
> is not important, even though it could affect the state machine in rare
> cases.

Do you have any examples of such cases?

The way I see it, if we have events from different files with the same 
timestamp, they would most probably be unrelated. If we do get 
"conflicting" information (a file open and a file close on the same 
file, by the same process, at the same time), this would probably be a 
problem with the kernel or more likely the tracing mechanism.


Alexandre


> Ideally, the state machine should become clever enough to detect
> these cases where relative ordering of events is unknown based on
> timestamps delta being under a certain threshold, or based on other
> instrumentat (interrupt execution surrounding the event). I'm not saying
> that LTTV and TMF are doing a particularly good at this at the moment --
> it should be considered as something to improve.
>
> Thanks,
>
> Mathieu
>