Paul, Thanks. We will look at that document and also check with the MVAPICH2 folks since it should be very similar. Rajeev
-----Original Message----- From: Paul H. Hargrove [mailto:[email protected]] Sent: Friday, August 29, 2008 3:14 PM To: Rajeev Thakur Cc: [email protected]; [email protected] Subject: Re: MPICH2 and BLCR
Rajeev,
Eric and I have not personally ever done the MPI work and thus are not able to give an estimate of the amount of work required. The only "documentation" on the subject I am aware of is the LAM/MPI paper about their work: http://ftg.lbl.gov/CheckpointRestart/Pubs/lacsi-2003.pdf I know OSU published something on their MVAPICH2+BLCR, but I don't recall any technical details being included.
However, both OSU and Cray have done integration of their respective MPICH2-derived MPIs with BLCR. So, I'd recommend contacting those groups for any guidance they may have as to time required or methodology. Let me know if you need contact info for either group. Also, the recent work from the MPICH-V group is using BLCR + MPICH2, though I don't have contacts there.
Finally, I can also connect you to the grad student at Indiana University who did the OpenMPI integration with BLCR, though that experience may be less directly relevant to you than the other 3.
One the work is underway, we'd be happy to answer any BLCR-related questions you may have, keeping in mind that we are NOT experts in implementing MPI.
-Paul
Rajeev Thakur wrote:
Hi Paul and Eric, We have been thinking about adding support for BLCR in MPICH2, and Intel also recently expressed interest in it for Intel MPI, which is based on MPICH2. Do you have an estimate for how much effort is involved in doing that? Is there any document or web page we can look at for more information on what needs to be done?
Thanks, Rajeev
-- Paul H. Hargrove [email protected] Future Technologies Group HPC Research Department Tel: +1-510-495-2352 Lawrence Berkeley National Laboratory Fax: +1-510-486-6900