I use MOAB with PETSc (and DMMoab) and build it as PETSc external package (via '--download-moab=1'). This works fine, however, not all MOAB tools are built in the process. Unfortunately, the Zoltan external package is broken and as a result mbpart is not being built. Could someone clarify the Zoltan dependence of mbpart and if there's a way to build it with just Parmetis? Thanks for your help. G.
Gerd,
I use MOAB with PETSc (and DMMoab) and build it as PETSc external package (via '--download-moab=1'). This works fine, however, not all MOAB tools are built in the process. Unfortunately, the Zoltan external package is broken and as a result mbpart is not being built.
The PETSc configuration with --download-moab currently only uses HDF5 and NetCDF is PETSc has been configured with these packages also through either --download-PACKAGE or --with-PACKAGE-DIR semantics. I don't believe there is currently a configuration setting that accomodates the Zoltan linkage through this workflow. You could however configure/compile MOAB with HDF5/NetCDF/Zoltan separately, install that somewhere and specify --with-moab-dir=/PATH/TO/MOAB. This is what I usually do since our development efforts use feature branches in both MOAB and PETSc simultaneously.
Could someone clarify the Zoltan dependence of mbpart and if there's a way to build it with just Parmetis?
Currently, mbpart depends on Zoltan interfaces to link against Metis/Parmetis and it is one of our top priority items to directly create an interface to at least Metis in the coming month so that building Zoltan for mbpart is not a requirement. Unfortunately until then, there isn't an elegant way to make this happen. Vijay On Fri, Mar 27, 2015 at 11:33 AM, Gerd Heber <[email protected]> wrote:
I use MOAB with PETSc (and DMMoab) and build it as PETSc external package (via '--download-moab=1'). This works fine, however, not all MOAB tools are built in the process. Unfortunately, the Zoltan external package is broken and as a result mbpart is not being built. Could someone clarify the Zoltan dependence of mbpart and if there's a way to build it with just Parmetis?
Thanks for your help. G.
Hi all, this could be a starting point. It’s similar to mbpart but based on METIS. The Makefile is very rudimental, the correct paths should be set manually!
On 27 Mar 2015, at 18:29, Vijay S. Mahadevan <[email protected]> wrote:
Gerd,
I use MOAB with PETSc (and DMMoab) and build it as PETSc external package (via '--download-moab=1'). This works fine, however, not all MOAB tools are built in the process. Unfortunately, the Zoltan external package is broken and as a result mbpart is not being built.
The PETSc configuration with --download-moab currently only uses HDF5 and NetCDF is PETSc has been configured with these packages also through either --download-PACKAGE or --with-PACKAGE-DIR semantics. I don't believe there is currently a configuration setting that accomodates the Zoltan linkage through this workflow.
You could however configure/compile MOAB with HDF5/NetCDF/Zoltan separately, install that somewhere and specify --with-moab-dir=/PATH/TO/MOAB. This is what I usually do since our development efforts use feature branches in both MOAB and PETSc simultaneously.
Could someone clarify the Zoltan dependence of mbpart and if there's a way to build it with just Parmetis?
Currently, mbpart depends on Zoltan interfaces to link against Metis/Parmetis and it is one of our top priority items to directly create an interface to at least Metis in the coming month so that building Zoltan for mbpart is not a requirement. Unfortunately until then, there isn't an elegant way to make this happen.
Vijay
On Fri, Mar 27, 2015 at 11:33 AM, Gerd Heber <[email protected]> wrote:
I use MOAB with PETSc (and DMMoab) and build it as PETSc external package (via '--download-moab=1'). This works fine, however, not all MOAB tools are built in the process. Unfortunately, the Zoltan external package is broken and as a result mbpart is not being built. Could someone clarify the Zoltan dependence of mbpart and if there's a way to build it with just Parmetis?
Thanks for your help. G.
Hi Lorenzo, Thanks much for the sources and this will certainly provide a good base to build upon for our metis interface. I will review/test this and create a PR so that we can make it part of the working repo at the earliest. Gerd, You can use the mbmetis tool as standalone replacement until mbpart is linked with metis directly and available on MOAB master. Please let us know if you find any inconsistencies in the partitioning interface or implementation. Thanks. Vijay On Fri, Mar 27, 2015 at 12:47 PM, Lorenzo Alessio Botti <[email protected]> wrote:
Hi all, this could be a starting point. It's similar to mbpart but based on METIS. The Makefile is very rudimental, the correct paths should be set manually!
On 27 Mar 2015, at 18:29, Vijay S. Mahadevan <[email protected]> wrote:
Gerd,
I use MOAB with PETSc (and DMMoab) and build it as PETSc external package (via '--download-moab=1'). This works fine, however, not all MOAB tools are built in the process. Unfortunately, the Zoltan external package is broken and as a result mbpart is not being built.
The PETSc configuration with --download-moab currently only uses HDF5 and NetCDF is PETSc has been configured with these packages also through either --download-PACKAGE or --with-PACKAGE-DIR semantics. I don't believe there is currently a configuration setting that accomodates the Zoltan linkage through this workflow.
You could however configure/compile MOAB with HDF5/NetCDF/Zoltan separately, install that somewhere and specify --with-moab-dir=/PATH/TO/MOAB. This is what I usually do since our development efforts use feature branches in both MOAB and PETSc simultaneously.
Could someone clarify the Zoltan dependence of mbpart and if there's a way to build it with just Parmetis?
Currently, mbpart depends on Zoltan interfaces to link against Metis/Parmetis and it is one of our top priority items to directly create an interface to at least Metis in the coming month so that building Zoltan for mbpart is not a requirement. Unfortunately until then, there isn't an elegant way to make this happen.
Vijay
On Fri, Mar 27, 2015 at 11:33 AM, Gerd Heber <[email protected]> wrote:
I use MOAB with PETSc (and DMMoab) and build it as PETSc external package (via '--download-moab=1'). This works fine, however, not all MOAB tools are built in the process. Unfortunately, the Zoltan external package is broken and as a result mbpart is not being built. Could someone clarify the Zoltan dependence of mbpart and if there's a way to build it with just Parmetis?
Thanks for your help. G.
I'll check out mbmetis for sure. For myself, I keep several versions of MOAB including one build with Zoltan. I'm really more concerned about my users who (rightly) roll their eyes every time I tell them they need to download and build packages A, B, C, ..., U, V, W, and also have multiple versions of P, Q and R on hand. I'm not trying to blame anybody, just trying to be helpful to my users. Have a great weekend. G. -----Original Message----- From: Vijay S. Mahadevan [mailto:[email protected]] Sent: Friday, March 27, 2015 1:35 PM To: Lorenzo Alessio Botti Cc: Gerd Heber; MOAB dev Subject: Re: [MOAB-dev] PETSc/MOAB/mbpart Hi Lorenzo, Thanks much for the sources and this will certainly provide a good base to build upon for our metis interface. I will review/test this and create a PR so that we can make it part of the working repo at the earliest. Gerd, You can use the mbmetis tool as standalone replacement until mbpart is linked with metis directly and available on MOAB master. Please let us know if you find any inconsistencies in the partitioning interface or implementation. Thanks. Vijay On Fri, Mar 27, 2015 at 12:47 PM, Lorenzo Alessio Botti <[email protected]> wrote:
Hi all, this could be a starting point. It's similar to mbpart but based on METIS. The Makefile is very rudimental, the correct paths should be set manually!
On 27 Mar 2015, at 18:29, Vijay S. Mahadevan <[email protected]> wrote:
Gerd,
I use MOAB with PETSc (and DMMoab) and build it as PETSc external package (via '--download-moab=1'). This works fine, however, not all MOAB tools are built in the process. Unfortunately, the Zoltan external package is broken and as a result mbpart is not being built.
The PETSc configuration with --download-moab currently only uses HDF5 and NetCDF is PETSc has been configured with these packages also through either --download-PACKAGE or --with-PACKAGE-DIR semantics. I don't believe there is currently a configuration setting that accomodates the Zoltan linkage through this workflow.
You could however configure/compile MOAB with HDF5/NetCDF/Zoltan separately, install that somewhere and specify --with-moab-dir=/PATH/TO/MOAB. This is what I usually do since our development efforts use feature branches in both MOAB and PETSc simultaneously.
Could someone clarify the Zoltan dependence of mbpart and if there's a way to build it with just Parmetis?
Currently, mbpart depends on Zoltan interfaces to link against Metis/Parmetis and it is one of our top priority items to directly create an interface to at least Metis in the coming month so that building Zoltan for mbpart is not a requirement. Unfortunately until then, there isn't an elegant way to make this happen.
Vijay
On Fri, Mar 27, 2015 at 11:33 AM, Gerd Heber <[email protected]> wrote:
I use MOAB with PETSc (and DMMoab) and build it as PETSc external package (via '--download-moab=1'). This works fine, however, not all MOAB tools are built in the process. Unfortunately, the Zoltan external package is broken and as a result mbpart is not being built. Could someone clarify the Zoltan dependence of mbpart and if there's a way to build it with just Parmetis?
Thanks for your help. G.
I'll check out mbmetis for sure. For myself, I keep several versions of MOAB including one build with Zoltan. I'm really more concerned about my users who (rightly) roll their eyes every time I tell them they need to download and build packages A, B, C, ..., U, V, W, and also have multiple versions of P, Q and R on hand. I'm not trying to blame anybody, just trying to be helpful to my users.
We understand your concern Gerd. I think once we have the metis partitioner linkage, we can ask PETSc to be configured with metis + hdf5 to make MOAB implementation support online re-partitioning as needed. There is also a separate effort to support --download-* in MOAB to ease the painful dependency build since this comes up often with some of our other users too. I hope to have all of this included by the next release and so keep watching for PRs and feature branches in the repo. Enjoy the weekend. Vijay On Fri, Mar 27, 2015 at 3:34 PM, Gerd Heber <[email protected]> wrote:
I'll check out mbmetis for sure. For myself, I keep several versions of MOAB including one build with Zoltan. I'm really more concerned about my users who (rightly) roll their eyes every time I tell them they need to download and build packages A, B, C, ..., U, V, W, and also have multiple versions of P, Q and R on hand. I'm not trying to blame anybody, just trying to be helpful to my users.
Have a great weekend. G.
-----Original Message----- From: Vijay S. Mahadevan [mailto:[email protected]] Sent: Friday, March 27, 2015 1:35 PM To: Lorenzo Alessio Botti Cc: Gerd Heber; MOAB dev Subject: Re: [MOAB-dev] PETSc/MOAB/mbpart
Hi Lorenzo,
Thanks much for the sources and this will certainly provide a good base to build upon for our metis interface. I will review/test this and create a PR so that we can make it part of the working repo at the earliest.
Gerd,
You can use the mbmetis tool as standalone replacement until mbpart is linked with metis directly and available on MOAB master. Please let us know if you find any inconsistencies in the partitioning interface or implementation. Thanks.
Vijay
On Fri, Mar 27, 2015 at 12:47 PM, Lorenzo Alessio Botti <[email protected]> wrote:
Hi all, this could be a starting point. It's similar to mbpart but based on METIS. The Makefile is very rudimental, the correct paths should be set manually!
On 27 Mar 2015, at 18:29, Vijay S. Mahadevan <[email protected]> wrote:
Gerd,
I use MOAB with PETSc (and DMMoab) and build it as PETSc external package (via '--download-moab=1'). This works fine, however, not all MOAB tools are built in the process. Unfortunately, the Zoltan external package is broken and as a result mbpart is not being built.
The PETSc configuration with --download-moab currently only uses HDF5 and NetCDF is PETSc has been configured with these packages also through either --download-PACKAGE or --with-PACKAGE-DIR semantics. I don't believe there is currently a configuration setting that accomodates the Zoltan linkage through this workflow.
You could however configure/compile MOAB with HDF5/NetCDF/Zoltan separately, install that somewhere and specify --with-moab-dir=/PATH/TO/MOAB. This is what I usually do since our development efforts use feature branches in both MOAB and PETSc simultaneously.
Could someone clarify the Zoltan dependence of mbpart and if there's a way to build it with just Parmetis?
Currently, mbpart depends on Zoltan interfaces to link against Metis/Parmetis and it is one of our top priority items to directly create an interface to at least Metis in the coming month so that building Zoltan for mbpart is not a requirement. Unfortunately until then, there isn't an elegant way to make this happen.
Vijay
On Fri, Mar 27, 2015 at 11:33 AM, Gerd Heber <[email protected]> wrote:
I use MOAB with PETSc (and DMMoab) and build it as PETSc external package (via '--download-moab=1'). This works fine, however, not all MOAB tools are built in the process. Unfortunately, the Zoltan external package is broken and as a result mbpart is not being built. Could someone clarify the Zoltan dependence of mbpart and if there's a way to build it with just Parmetis?
Thanks for your help. G.
participants (3)
-
Gerd Heber -
Lorenzo Alessio Botti -
Vijay S. Mahadevan