[mpich-discuss] Coordinated Checkpoint without making checkpoint images

Darius Buntinas buntinas at mcs.anl.gov
Wed Nov 9 10:54:07 CST 2011


If you're not taking a checkpoint image, you don't need blcr.  blcr is only used to take an image of an individual process.  So if all you want to do is do the distributed checkpointing protocol, then you don't need blcr at all.

-d

On Nov 9, 2011, at 10:07 AM, Mohammed El Mehdi DIOURI wrote:

> Hi Darius,
> 
> Thanks for your answer.
> 
> In fact, in my configuration, mpich is set with blcr checkpoint lib.
> I don't need to modify something in the blcr code ?
> 
> All I need to do is in the MPIDI_nem_ckpt_finish() function of mpid_nem_ckpt.c of mpich ?
> 
> Thank you for your help,
> 
> Mehdi.
> 
> 
> Le 1 nov. 2011 à 20:20, Darius Buntinas a écrit :
> 
>> Hi Mehdi,
>> 
>> You'll need to modify the MPIDI_nem_ckpt_finish() function in mpid_nem_ckpt.c.  This function is called when the checkpoint protocol has completed and we're ready to take a checkpoint of the process.  The checkpoint is taken between the sem_post and sem_wait (the blcr checkpoint thread (i.e., the ckpt_cb function) is waiting on the ckpt_sem to take the checkpoint).  If you're doing this without blcr, you'll also need to modify the mechanism to initiate the checkpoint.  Normally the checkpoint thread will set MPIDI_nem_ckpt_start_checkpoint = TRUE at rank 0, so you'll need to do that yourself.
>> 
>> -d
>> 
>> 
>> On Oct 27, 2011, at 2:33 PM, Mohammed El Mehdi DIOURI wrote:
>> 
>>> Hi,
>>> 
>>> I was wondering if with mpich2 we can run the coordinated checkpointing without making the checkpoint images. 
>>> I mean for each checkpoint interval, we only play the messages that enable the corresponding coordination ?
>>> 
>>> Thanks for your help,
>>> 
>>> Mehdi.
>>> _______________________________________________
>>> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
>>> To manage subscription options or unsubscribe:
>>> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
>> 
>> _______________________________________________
>> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
>> To manage subscription options or unsubscribe:
>> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
> 



More information about the mpich-discuss mailing list