On Thu, Dec 23, 2010 at 7:46 AM, Mark F. Adams <adams at pppl.gov> wrote:
> What other games are there for scaleable partitioners? > > Scotch looks interesting from their web page, does anyone have any > experience with it? They seem to claim 2K processor runs which is about one > or two orders short of showing that it might scale. > > N.B. I've tried Chaco but found it was not even parallel. Also, I do not know if any other package does multiobjective partitioning like ParMetis does. Matt > > Mark > > > On Dec 23, 2010, at 8:44 AM, Barry Smith wrote: > > >> On Dec 22, 2010, at 9:58 PM, Mark F. Adams wrote: >> >> I've been concerned for a while that, as Barry says, ParMetis seems >>> orphaned. I've had problems with scaling ParMetis up on BG for a while. I >>> believe they allocate a PxP (integer) matrix which kills my code (my code is >>> flat MPI but I make sub communicators so that each ParMetis MPI process has >>> the whole node to itself). Can anyone scale Parmetis up on BG-P? >>> >>> Anyway this has been on my wish list for a while so I thought I'd use >>> this opportunity to see if anyone has ideas for a solution. >>> >> >> Switch to a non-orphaned partitioning package. Just like Microsoft >> Windows isn't the only game in town, ParMetis isn't the only game in town. >> >> Barry >> >> >>> Mark >>> >>> On Dec 22, 2010, at 7:50 PM, Matthew Knepley wrote: >>> >>> On Wed, Dec 22, 2010 at 3:02 PM, Lisandro Dalcin <dalcinl at gmail.com> >>>> wrote: >>>> On 22 December 2010 19:57, Barry Smith <bsmith at mcs.anl.gov> wrote: >>>> >>>>> >>>>> On Dec 22, 2010, at 4:51 PM, Lisandro Dalcin wrote: >>>>> >>>>> On 21 December 2010 20:50, Sean Farley <sean at mcs.anl.gov> wrote: >>>>>> >>>>>>> fixing/understanding 64-bit checks, >>>>>>>>> >>>>>>>> >>>>>>>> I think these checks are ok, please report problems. >>>>>>>> >>>>>>> >>>>>>> Would it be possible to get ParMetis working with 64bit indices? >>>>>>> Sean >>>>>>> >>>>>> >>>>>> This is from parmetis.h >>>>>> >>>>>> /* Indexes are as long as integers for now */ >>>>>> #ifdef IDXTYPE_INT >>>>>> typedef int idxtype; >>>>>> #else >>>>>> typedef short idxtype; >>>>>> #endif >>>>>> >>>>>> Have any of you seriously tryied parmetis in 64 bits? >>>>>> >>>>> >>>>> Yes, when I noticed that block of code several years ago. Looks like it >>>>> would be easy to switch to long int but sadly this include is misleading >>>>> about the simplicity, turns out this typedef beasty is not used throughout >>>>> the source code consistently. >>>>> >>>> >>>> Damn, >>>> >>>> I sent email about the 64 bit int at that time and got back the >>>>> response "in a few months". Note that his non-MPI codes may fully support >>>>> the 64 bit (I'm guessing that is where the $ is so is all he cares about). >>>>> That was like two years ago. I consider Parmetis orphaned, not >>>>> maintained, >>>>> hence in the HPC world, dead. >>>>> >>>>> >>>> Add to that license issues. A fork in order to maintain and improve it >>>> is unlikely. >>>> >>>> If this is a big deal, I can mail George directly. He let me fool with >>>> the tarball and distribute it >>>> with PETSc last time I asked him. >>>> >>>> Matt >>>> >>>> -- >>>> Lisandro Dalcin >>>> --------------- >>>> CIMEC (INTEC/CONICET-UNL) >>>> Predio CONICET-Santa Fe >>>> Colectora RN 168 Km 472, Paraje El Pozo >>>> Tel: +54-342-4511594 (ext 1011) >>>> Tel/Fax: +54-342-4511169 >>>> >>>> >>>> >>>> -- >>>> What most experimenters take for granted before they begin their >>>> experiments is infinitely more interesting than any results to which their >>>> experiments lead. >>>> -- Norbert Wiener >>>> >>> >>> >> >> > -- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20101223/204c0343/attachment.html>
