Daniel,

For very high-rate things across many channels, what Blaise is talking
about might lead to problems.  However, I don't think it is in this
particular case.

I ran your tests and checked with rxbag and the packets seem to be
nicely distributed as we would want.

I am 99% confident that this is a problem with rosbag in boxturtle.
The old rosbag (which actually called through to rosplay), had a lot
of problems in the timing engine that led to this kind of bursty
behavior on playback.  For cturtle, I rewrote the rosbag timing engine
and I believe it is now far more robust and well-behaved.

I just confirmed playback from boxturtle only invoked the callback 54
times.  Playback on boxturtle invoked the callback 2494 times.

If you can possibly move to cturtle, I believe that should solve your
issues.  If you really need to work with boxturtle I can investigate
the possibility of patching in some of these timing fixes.

--Jeremy

On Fri, Jul 30, 2010 at 10:03 AM, Daniel Grollman
<daniel.grollman@epfl.ch> wrote:
> Hi Blaise,
>
>        Thanks for this info, I hadn't thought to consider the scheduler.
> However, opening up my bagfile with rxbag (and zooming in very close),
> it looks like there is indeed a packet every 2ms.  I'll try to confirm
> this with the API.
>
> Dan
>
> On 07/30/2010 06:15 PM, Blaise Gassend wrote:
>> As far as I know, rosbag can only rely on message arrival times when it
>> records a data stream. At 500 Hz, your messages are being sent at a rate
>> that is fast compared with Linux's scheduling granularity, so on a
>> loaded system, I would not expect rosbag to necessarily be scheduled
>> each time it receives a message. Hence it will receive the messages in
>> clumps, and during replay, messages will be played back in clumps. You
>> can verify whether this is the case by looking at the timestamps in the
>> bag. I think rxbag should allow you to view the timestamps graphically.
>> The python bag API should allow you to view the timestamps directly.
>>
>> You might be able to improve things by puting rosbag on one of the
>> "realtime" queues using schedtool.
>>
>> You can also see the scheduling happening using the kernel ftrace
>> mechanism.
>>
>> On Fri, 2010-07-30 at 14:02 +0200, Daniel Grollman wrote:
>>> Hello,
>>>
>>>      I've noticed some odd behavior with rosbag.  I'm generating data at
>>> 500hz, and recording a bag.  However, when playing back the bag, only a
>>> fraction of the messages reach my callback.  I believe that rosbag is
>>> 'bursting' the messages, i.e., sporadically sending them at much faster
>>> than 500 Hz, so they are then getting lost as my queue size is only 1
>>> (which is what I want).
>>>
>>> I'm running ros 1.0.4, and ubuntu 9.10.  I've put together a little demo
>>> of this issue, code is here:
>>>
>>> http://lasa.epfl.ch/~dang/rosbag_test.tar.gz
>>>
>>> Thank you,
>>>
>>> Dan Grollman
>>> EPFL
>>> _______________________________________________
>>> ros-users mailing list
>>> ros-users@code.ros.org
>>> https://code.ros.org/mailman/listinfo/ros-users
>>
>>
>> _______________________________________________
>> ros-users mailing list
>> ros-users@code.ros.org
>> https://code.ros.org/mailman/listinfo/ros-users
> _______________________________________________
> ros-users mailing list
> ros-users@code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users
>