0

I'm trying to use dmtcp for task migration. Running dmtcp_restart on the same computer works correctly. However, after the generated checkpoint file and related input/output files were transferred to the target computer, dmtcp_restart gives segmentation fault. Does dmtcp support task migration? If so, could anybody please suggest the solution (or other tools)?

Thank you Best regards.

Udomo
  • 25
  • 4

2 Answers2

1

Well I've also encountered the same issue with dmtcp, though I know a person who claims that it is possible. Anyways since you asked for other tools, BLCR does work pretty well for process migration, also the checkpoint image by blcr is of smaller size. Try using it.

addu
  • 175
  • 8
1

DMTCP does support process migration. One possible way to ensure a successful migration is to specify the --ckpt-open-files flag to dmtcp_launch and then transfer the ckpt images and the corresponding directories to the remote node.

If you are using homogeneous hosts, everything should work smoothly. However, if you are using heterogeneous systems, the result may vary. Ideally, DMTCP should be able to migrate process within different kernel versions as long as the same binary can run on both. There can be issues if the kernel version are distant and have incompatibilities. In any case, if process migration doesn't work for your application, feel free to write to us at dmtcp-forum@lists.sourceforge.net or open a github issue at https://github.com/dmtcp/dmtcp and we'll try to resolve your issue as soon as possible.

Kapil Arya
  • 155
  • 8