From: Paul H. Hargrove (PHHargrove_at_lbl_dot_gov)
Date: Tue Nov 10 2009 - 14:02:50 PST
Chris, Currently BLCR will exclude : 1) Pages mapped from a file and unmodified, which will include the code and read-only data segments of the executable and shared libraries. This may also included mmap()s performed by the application, if any. 2) "Anonymous" (non file-backed) pages that are mapped in to the process but not yet written to, which includes unused portions of the stack and heap. In the next release (target Jan 2010) we will have a mechanism that allows applications and libraries to hint to BLCR about regions of memory that can be excluded. For instance, the OpenMPI developer that works on checkpoint/restart has suggested that they might exclude any of MPI's receive buffers that don't hold actual messages at the time the checkpoint is taken. We are also working on incremental checkpointing for the next release. In the case where one takes multiple checkpoints of the same process over time, this will omit from a checkpoint any pages not written to since the previous one. However, when using this mechanism one needs all the past checkpoints to restore (and so more disk space is required). -Paul Chris Chaudoir wrote: > To Paul et all, > > I just started playing around with the latest version of BLCR (nice > btw) and had a question...besides the expected methods of skipping the > code and unused stack, can you describe the 'memory exclusion' methods > in BLCR (or point me to the right documentation)? > > > Respectfully, > Chris Chaudoir > > BSEE & CS grad student > -- Paul H. Hargrove PHHargrove_at_lbl_dot_gov Future Technologies Group Tel: +1-510-495-2352 HPC Research Department Fax: +1-510-486-6900 Lawrence Berkeley National Laboratory