Hello,
I hope we have the right mailing list. We are two of a team of
developers for DMTCP (transparent checkpoint-restart). It runs under
Linux, and is free software (LGPLv3+), found at:
http://dmtcp.sourceforge.net
In our latest release, version 1.2.3, we have done some simple
testing on R, and find that we can successfully checkpoint and restart.
Its usage can be as simple as:
dmtcp_checkpoint R [ run under checkpoint control ]
dmtcp_command --checkpoint [ from a diff. window, creates ckpt_R_*.dmtcp ]
dmtcp_restart ckpt_R_*.dmtcp [ from original window ]
ALTERNATIVELY: ./dmtcp_restart_script.sh
We are interested in supporting R to a fuller extent. But we don't
have enough experience with R to do more thorough testing. We have tested
DMTCP on a small test in C, which frequently loads and unloads libraries
(w/ dlopen). We added this test in C, because we understand that loading
and unloading libraries is part of the core design of R.
Would it be possible for someone to try out R/DMTCP on some realistic
R programs and give us feedback on what works or doesn't work? We believe
that transparent checkpointing would be helpful to the R community for
any long-running programs.
Thanks and best wishes,
- Gene and Kapil (representing the DMTCP team)
