The Dynamite checkpointer 2.0

This checkpointer enables the user to transparently add checkpointing facilities to his application, i.e., no modifications to either the user's program source code, nor to the kernel are needed. This checkpointer is available under GNU GPL.

Known limitations:

• Programs using pipes or sockets for communication are not supported
• Programs using signals for synchronization are not supported
• Programs using threads are not supported
• Programs using libc with TLS option are not supported (normal non-TLS library should be used)

Downloading your copy of the checkpointer

In order to download a copy of Dynamite Checkpointer please check either source or binary directory. There you may find various versions of the software, depending on glibc package version you want to use. Some of the releases are suited for RedHat-modified glibc library. After package name there is an exact name and version of the glibc library compatible with this particular release. After two subsequent underscores there is an actual version of the checkpointer release.

Please note that our checkpointer does not support any cases of threaded programs whatever thread library is used. In case of glibc-2.3.2 when it is compiled with enabled NPTL (thread library) and TLS (Thread-Local Storage) options the job has to be linked with traditional, non-TLS version of the libc library. Vendors usually provide that along with the default version. A compilation option in Makefile like:
-L/lib -Wl,-rpath,/lib
should make it work (here /lib is the location of non-TLS version of libc.so.6 library).

Documentation

The documentation on internals of the checkpointer can be found inside the package distribution as well as a brief note on installation and usage of the provided software. The essentials are duplicated here for your convenience.

Install:

• Download the source files (in src/elf/).
• Get the glibc sources appropriate for your distribution (you cannot compile the dynamic loader w/o glibc sources).
• Copy the src/elf/ files to glibc's elf/ subdirectory (overwriting Makefile and rtld.c).
• Compile the whole glibc (see the installation notes in glibc package for guidelines). Should you already have a glibc build tree for some reason, typing make in the elf/ subdirectory should be enough.
• Copy ld.so, your new dynamic loader with checkpointing support, to a convenient location.

Use:

• Pass the following options on the compiler's commandline during the final linking of your executable:
  -Wl,-dynamic-linker,/path/to/dynamic/loader/ld.so -rdynamic
• Send USR1 signal to a running process to checkpoint it:
  kill -USR1 pid
• The checkpoint file is a standalone executable: simply execute it like any other program to restore your job.

The page of the old checkpointer also contains some useful information.

Contact person

Breanndán Ó Nualláin (mail: bon AT science DOT uva DOT nl)

Most recent modification: Tue 7 September, 2004 by Tomasz Gubała