Hi Edson,
Kapil and I were tlaking together just now.
The best thing still seems to be to wait for version 2.4.2 (which should
be available in a few days). In that version:
--interval XXX --no-coordinator will work.
[ --no-coordinator still needs to be fixed in the development branch.
We'll do that for the 2.4.2 release. ]
(Hopefully, I understood this time what you're requesting. :-) )
Best,
- Gene and Kapil
On Thu, Oct 08, 2015 at 11:42:45PM +0200, Edson Tavares de Camargo wrote:
> Hi Gene!
>
> Thank you for your reply.
>
> I know about that bug on DMTCP_CHECKPOINT_INTERVAL, version 2.4.1. My
> previous question are about the version 2.2.
>
> When I call ~/test/plugin/applic-initiated-ckpt/applic$ dmtcp_launch
> --no-coordinator ./applic the applic, should the applic wait for the value
> defined on DMTCP_CHECKPOINT_INTERVAL? (DMTCP version 2.2)
>
> In the version 2.4.0 the --no-coordinator option are not working.
>
> Thanks!
>
> Edson
> On Oct 8, 2015 10:20 PM, "Gene Cooperman" <[email protected]> wrote:
>
> > Hi Edson,
> > You seem to have hit a known bug that we have for dmtcp version 2.4.1.
> > We had an unfortunate regression concerning interval checkpointing.
> > --interval and DMTCP_CHECKPOINT_INTERVAL are not working properly
> > in version 2.4.1.
> >
> > We will be releasing version 2.4.2 in a few days. In the meantime,
> > your options are to use dmtcp-2.4.0, or else the development
> > branch (which is currently reasonable stable). The development branch
> > can be found through:
> > google dmtcp download
> > --> http://dmtcp.sourceforge.net/downloads.html
> > ----> git clone https://github.com/dmtcp/dmtcp.git
> > ----> OR: wget https://github.com/dmtcp/dmtcp/archive/master.zip
> >
> > Best,
> > - Gene
> >
> > On Thu, Oct 08, 2015 at 07:29:38PM +0200, Edson Tavares de Camargo wrote:
> > > Hi Kapil, I will comment below:
> > >
> > > 2015-10-08 15:39 GMT+02:00 Kapil Arya <[email protected]>:
> > >
> > > > Hi Edson,
> > > >
> > > > For coordinator-less checkpointing, I would suggest that you use the
> > > > "--no-coordinator" flag with dmtcp_launch.
> > > >
> > >
> > > The version 2.2 woks fine with --no-coordination:
> > >
> > > ~/test/plugin/applic-initiated-ckpt/applic$ dmtcp_launch
> > --no-coordinator
> > > ./applic
> > >
> > >
> > >
> > > > This allows you to specify an checkpoint interval.
> > > >
> > >
> > > In this case above, how the checkpoint interval works? Should the
> > applic.c
> > > wait until DMTCP_CHECKPOINT_INTERVAL for makes the checkpoint. I ask
> > > because seems that it not waiting for DMTCP_CHECKPOINT_INTERVAL.
> > >
> > >
> > > > Further, you can also provide a port number with "--port" and then use
> > > > dmtcp_command to request checkpoints explicitly.
> > > >
> > >
> > > I would like that each process start the checkpoint in a different
> > > interval. For sample, suppose 4 MPI processes:
> > > - process 0 makes checkpoint every 5 seconds
> > > - process 1 makes checkpoint every 8 seconds
> > > - process 2 makes checkpoint every 3 seconds
> > > so on...
> > >
> > > Can I set that behaviour, that is, both interval and request checkpoints
> > > directly in my MPI application code?
> > >
> > >
> > > Thank you again!
> > >
> > > Edson
> >
> > >
> > ------------------------------------------------------------------------------
> >
> > > _______________________________________________
> > > Dmtcp-forum mailing list
> > > [email protected]
> > > https://lists.sourceforge.net/lists/listinfo/dmtcp-forum
> >
> >
------------------------------------------------------------------------------
_______________________________________________
Dmtcp-forum mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dmtcp-forum
