- mpiwg-ft - mailman3.cels.anl.gov

Re: [Mpi3-ft] Recoverability of MPI State Proposal Discussion
by Richard Graham 06 Mar '08

06 Mar '08

After thinking about this a bit, I decided to take Josh¹s suggestion one more step. The proposal attached to the UTK paper is a good starting point for this whole discussion, so I obtained the permission of the primary authors to use this as a starting point. I will upload this to the web site later today (I hope). While this does not cover items such as dynamic communicators, and even the original authors would do things differently, it is a very good starting point. Rich On 3/3/08 9:50 AM, "Richard Graham" <rlgraham(a)ornl.gov> wrote: > I am no the hook to produce a first draft of a document for this. I am > planning to do this later in the week, after I finish the 2.1 document > reviews I need to do. > > Rich > > > On 3/3/08 8:45 AM, "Josh Hursey" <jjhursey(a)open-mpi.org> wrote: > >> > I'd like to start the discussion on the proposal for "defining >> > communicator state with process loss", or recoverability of MPI state >> > in the presence of process loss. >> > >> > To start the discussion I am suggesting that folks have a look at the >> > following paper since I think it relates to quite a number of related >> > topics: >> > >> > Fagg, G., et. al. "Extending the MPI Specification for Process Fault >> > Tolerance on High Performance Computing Systems" >> > http://icl.cs.utk.edu/projectsfiles/ftmpi/pubs/isc2004-FT-MPI.pdf >> > >> > ---- >> > Josh Hursey >> > Graduate Student, Indiana University >> > >> > >> > _______________________________________________ >> > Mpi3-ft mailing list >> > Mpi3-ft(a)lists.mpi-forum.org >> > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-ft > > _______________________________________________ > Mpi3-ft mailing list > Mpi3-ft(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-ft >

1 0

Re: [Mpi3-ft] Recoverability of MPI State Proposal Discussion
by Richard Graham 03 Mar '08

03 Mar '08

I am no the hook to produce a first draft of a document for this. I am planning to do this later in the week, after I finish the 2.1 document reviews I need to do. Rich On 3/3/08 8:45 AM, "Josh Hursey" <jjhursey(a)open-mpi.org> wrote: > I'd like to start the discussion on the proposal for "defining > communicator state with process loss", or recoverability of MPI state > in the presence of process loss. > > To start the discussion I am suggesting that folks have a look at the > following paper since I think it relates to quite a number of related > topics: > > Fagg, G., et. al. "Extending the MPI Specification for Process Fault > Tolerance on High Performance Computing Systems" > http://icl.cs.utk.edu/projectsfiles/ftmpi/pubs/isc2004-FT-MPI.pdf > > ---- > Josh Hursey > Graduate Student, Indiana University > > > _______________________________________________ > Mpi3-ft mailing list > Mpi3-ft(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-ft

1 0

[Mpi3-ft] Recoverability of MPI State Proposal Discussion
by Josh Hursey 03 Mar '08

03 Mar '08

I'd like to start the discussion on the proposal for "defining communicator state with process loss", or recoverability of MPI state in the presence of process loss. To start the discussion I am suggesting that folks have a look at the following paper since I think it relates to quite a number of related topics: Fagg, G., et. al. "Extending the MPI Specification for Process Fault Tolerance on High Performance Computing Systems" http://icl.cs.utk.edu/projectsfiles/ftmpi/pubs/isc2004-FT-MPI.pdf ---- Josh Hursey Graduate Student, Indiana University

1 0

Re: [Mpi3-ft] Transactional Messages
by Josh Hursey 02 Mar '08

02 Mar '08

On Mar 1, 2008, at 7:36 PM, Greg Bronevetsky wrote: > >> I took the liberty of converting the 2 txt files to wiki pages. This >> will allow us to better use the features of the wiki, like change >> logs. They are linked off of the main FT wiki page: >> http://svn.mpi-forum.org/trac/mpi-forum-web/wiki/ >> FaultToleranceWikiPage > How did you add the files as wiki pages? I see how to attach > documents but I don't see how to add pages. To add a page to the wiki: - click on the "Edit this page" button at the bottom of the page on which you want to add the link to the page. So typically the mail FaultToleranceWikiPage. - Whereever you want the link to appear just add the text: [wiki:"Transactional-Msg" Transactional Messages] Just after the "wiki:" this will be the page title to show up in the url. After that is the link text to show up on the webpage as the link. - Save this page with the "submit changes" button. You may want to Preview the page before submitting just to make sure the page will appear as you wanted it to. - When you return to the page you added the link to, just follow your new link. The wiki will have automatically created a stub page for you. - To edit the stub page just click on the "edit this page" button at the bottom of that new page. > >> I suggest that we remove the two .txt files from the wiki head so we >> do not become confused as to which document we are talking about. > Sounds good to me. Where's the button for removing the files? I was wondering the same thing. When you click on the attachment it will take you to an HTML page with the attachment. At the bottom of that page there should be a button "Delete attachment" For some reason this button does not appear with the FT Wiki. I wonder if some of the permissions on the wiki are messed up such that you can only delete a page if you are logged in. Just a guess though. -- Josh > > Greg Bronevetsky > Post-Doctoral Researcher > 1028 Building 451 > Lawrence Livermore National Lab > (925) 424-5756 > bronevetsky1(a)llnl.gov > _______________________________________________ > Mpi3-ft mailing list > Mpi3-ft(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-ft

1 0

Re: [Mpi3-ft] Transactional Messages
by Greg Bronevetsky 01 Mar '08

01 Mar '08

>I took the liberty of converting the 2 txt files to wiki pages. This >will allow us to better use the features of the wiki, like change >logs. They are linked off of the main FT wiki page: > http://svn.mpi-forum.org/trac/mpi-forum-web/wiki/ >FaultToleranceWikiPage How did you add the files as wiki pages? I see how to attach documents but I don't see how to add pages. >I suggest that we remove the two .txt files from the wiki head so we >do not become confused as to which document we are talking about. Sounds good to me. Where's the button for removing the files? Greg Bronevetsky Post-Doctoral Researcher 1028 Building 451 Lawrence Livermore National Lab (925) 424-5756 bronevetsky1(a)llnl.gov

1 0

[Mpi3-ft] Con call
by Richard Graham 27 Feb '08

27 Feb '08

Lets not have the call this week, and focus on getting done with writing up a first cut at the docs we signed up to start writing for our face-to-face meeting in 2 weeks. To date I think Greg is the only one that has done this, and there are a few more first drafts that need to be written. Any objections ? Rich

1 0

[Mpi3-ft] "Fault handling = error handling" proposal added
by Supalov, Alexander 27 Feb '08

27 Feb '08

Hi everybody, I've added to the Wiki a proposal that suggests reuse of the existing MPI error handling mechanism for fault notification (see http://svn.mpi-forum.org/trac/mpi-forum-web/wiki/Fault%20Handling) Best regards. Alexander -- Dr Alexander Supalov Intel GmbH Hermuelheimer Strasse 8a 50321 Bruehl, Germany Phone: +49 2232 209034 Mobile: +49 173 511 8735 Fax: +49 2232 209029 --------------------------------------------------------------------- Intel GmbH Dornacher Strasse 1 85622 Feldkirchen/Muenchen Germany Sitz der Gesellschaft: Feldkirchen bei Muenchen Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer Registergericht: Muenchen HRB 47456 Ust.-IdNr. VAT Registration No.: DE129385895 Citibank Frankfurt (BLZ 502 109 00) 600119052 This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.

1 0

Re: [Mpi3-ft] Transactional Messages
by Josh Hursey 25 Feb '08

25 Feb '08

On Feb 21, 2008, at 5:15 PM, Greg Bronevetsky wrote: > BTW, I don't know how to use > this Wiki, so I've just uploaded my .txt file as a page attachment. > If there's a cleaner way to do this, someone please let me know. I took the liberty of converting the 2 txt files to wiki pages. This will allow us to better use the features of the wiki, like change logs. They are linked off of the main FT wiki page: http://svn.mpi-forum.org/trac/mpi-forum-web/wiki/ FaultToleranceWikiPage If you click on the "Edit this page" button at the bottom of any Wiki page you can edit that page. Wiki formatting is quite simple once you get the basic feel of it. There is a whole page describing the particulars here: http://svn.mpi-forum.org/trac/mpi-forum-web/wiki/WikiFormatting It is many times easier to learn by example, so if you just follow the formatting examples in these pages it will show you basically all you need to know. I suggest that we remove the two .txt files from the wiki head so we do not become confused as to which document we are talking about. -- Josh

1 0

Re: [Mpi3-ft] Transactional Messages
by Greg Bronevetsky 25 Feb '08

25 Feb '08

>So are you really talking about adding a whole new set of pt2pt and >collectives APIs specifically to handle transactional communications? Or >are you looking to extend the current APIs to have a mode that a user >could set the error detection? I have to admit I still find the >difference between the two modes that Rich describes not that far apart >from each other. I don't think that we're that far ahead on this but my personal opinion is that we should not add new sets of MPI calls unless it is truly necessary. I think that in this case we can get all the functionality we need by adding - a new function that the user would call at initialization time to set the error detection/notification policy and - communicating error events via new return values and error handler calls. Greg Bronevetsky Post-Doctoral Researcher 1028 Building 451 Lawrence Livermore National Lab (925) 424-5756 bronevetsky1(a)llnl.gov

1 0

Re: [Mpi3-ft] Transactional Messages
by Terry Dontje 25 Feb '08

25 Feb '08

Greg Bronevetsky wrote: > At 09:38 AM 2/23/2008, Richard Graham wrote: >> So I think we are some what talking past each other. I think that >> what you really care about >> with respect to communications errors is information on messages that >> have not completed, >> and, more important, canï¿½t complete ? Is this correct ? >> I have focused more on errors that occur, but the low-level can >> handle them and does not need >> to pass information about them back up to the user. I believe this is >> where we said that we >> may want to be able to give the app some indication on performance >> degradation, at their >> request. Is this correct ? > > Exactly. My point was that the former belongs in the transactional > memory API, while the latter belongs in the QoS API. Kannan and I are > tasked with drafting something for the latter. One thing to note > though is that identical low-level events may fall under either API. > In particular, a single-bitflip may result in a message drop in one > MPI implementation and a seamless recovery with minor performance > degradation in another. The MPI implementation gets to choose the > mapping between low-level events and their high-level manifestations. > Right now we are just trying to define a reasonable interface for the > high-level manifestations. So are you really talking about adding a whole new set of pt2pt and collectives APIs specifically to handle transactional communications? Or are you looking to extend the current APIs to have a mode that a user could set the error detection? I have to admit I still find the difference between the two modes that Rich describes not that far apart from each other. --td

1 0