I don't have parallel partitions, I'm looking into HelloMOABPar example. Thanks. Rajeev ----- Original Message ----- From: "Grindeanu, Iulian R." <[email protected]> To: Vijay S. Mahadevan <[email protected]> Cc: "Jain, Rajeev" <[email protected]>; MOAB Dev <[email protected]> Sent: Thursday, July 31, 2014 10:12 AM Subject: RE: [MOAB-dev] 137GB mesh file mbconvert is parallel-capable mbpart is not mbsize "is" , but it needs improvements for output agree we need to review all tools ________________________________________ From: Vijay S. Mahadevan [[email protected]] Sent: Thursday, July 31, 2014 9:36 AM To: Grindeanu, Iulian R. Cc: Jain, Rajeev; MOAB Dev Subject: Re: [MOAB-dev] 137GB mesh file
I think we need to modify mbsize to run better in parallel (for example, to gather all counts from each processor, for the owned entities) and print only once , something like HelloParMOAB
Hmm ok. Make all the tools parallel capable then ? mbsize, mbpart, mbconvert etc. Iulian, we need to make a list of things TODO and then we can prioritize what's needed right now. I'll also get started on a list in parallel and we can consolidate. Vijay On Thu, Jul 31, 2014 at 9:30 AM, Grindeanu, Iulian R. <[email protected]> wrote:
Hi Rajeev, You can use mbsize -p0 option This will load the file in parallel, but the printout will be mangled between processors, each processor will write its count (-p1 will resolve shared entities, too), -p2 will exchange a layer of ghost elements)
so you can do something like this mpiexec -np 64 mbsize -p0 <file_name>
How many partitions do you have?
Do you have partitions? (parallel partitions?)
You can also modify HelloParMOAB example for you needs (harder)
LoadPartial example can load only a few partitions, and write a smaller file, it can be run in serial
Something like ./LoadPartial <file_name> PARALLEL_PARTITION 0 1 5
This will load only partitions 0, 1 and 5, in serial, and write a part.h5m file, for which you can do an mbsize in serial (if it is small enough)
All these methods will ensure at least that you have a "readable" file .
I think we need to modify mbsize to run better in parallel (for example, to gather all counts from each processor, for the owned entities) and print only once , something like HelloParMOAB
Iulian
________________________________ From: [email protected] [[email protected]] on behalf of Rajeev Jain [[email protected]] Sent: Wednesday, July 30, 2014 11:40 PM To: MOAB Dev Subject: [MOAB-dev] 137GB mesh file
I have an ABTR model (billions of hex elements, created by MeshKit/rgg) that is 137GB on fusion@ANL: /homes/jain/sigma/moab/lasso/meshkit/rgg/test/c_abtr_sh
I'm unable to do mbsize on it -> any suggestions? Just want to make sure if this is a valid file, also want to know material/boundary sets contained in it.
Rajeev
How did you generate it, with rgg? it will be tough to read it in parallel. Do you have materials sets on it? You can maybe use MATERIAL_SET tag to load partially, look at LoadPartial example. (you need to know the material values) ________________________________________ From: Rajeev Jain [[email protected]] Sent: Thursday, July 31, 2014 12:14 PM To: Grindeanu, Iulian R.; Vijay S. Mahadevan Cc: MOAB Dev Subject: Re: [MOAB-dev] 137GB mesh file I don't have parallel partitions, I'm looking into HelloMOABPar example. Thanks. Rajeev ----- Original Message ----- From: "Grindeanu, Iulian R." <[email protected]> To: Vijay S. Mahadevan <[email protected]> Cc: "Jain, Rajeev" <[email protected]>; MOAB Dev <[email protected]> Sent: Thursday, July 31, 2014 10:12 AM Subject: RE: [MOAB-dev] 137GB mesh file mbconvert is parallel-capable mbpart is not mbsize "is" , but it needs improvements for output agree we need to review all tools ________________________________________ From: Vijay S. Mahadevan [[email protected]] Sent: Thursday, July 31, 2014 9:36 AM To: Grindeanu, Iulian R. Cc: Jain, Rajeev; MOAB Dev Subject: Re: [MOAB-dev] 137GB mesh file
I think we need to modify mbsize to run better in parallel (for example, to gather all counts from each processor, for the owned entities) and print only once , something like HelloParMOAB
Hmm ok. Make all the tools parallel capable then ? mbsize, mbpart, mbconvert etc. Iulian, we need to make a list of things TODO and then we can prioritize what's needed right now. I'll also get started on a list in parallel and we can consolidate. Vijay On Thu, Jul 31, 2014 at 9:30 AM, Grindeanu, Iulian R. <[email protected]> wrote:
Hi Rajeev, You can use mbsize -p0 option This will load the file in parallel, but the printout will be mangled between processors, each processor will write its count (-p1 will resolve shared entities, too), -p2 will exchange a layer of ghost elements)
so you can do something like this mpiexec -np 64 mbsize -p0 <file_name>
How many partitions do you have?
Do you have partitions? (parallel partitions?)
You can also modify HelloParMOAB example for you needs (harder)
LoadPartial example can load only a few partitions, and write a smaller file, it can be run in serial
Something like ./LoadPartial <file_name> PARALLEL_PARTITION 0 1 5
This will load only partitions 0, 1 and 5, in serial, and write a part.h5m file, for which you can do an mbsize in serial (if it is small enough)
All these methods will ensure at least that you have a "readable" file .
I think we need to modify mbsize to run better in parallel (for example, to gather all counts from each processor, for the owned entities) and print only once , something like HelloParMOAB
Iulian
________________________________ From: [email protected] [[email protected]] on behalf of Rajeev Jain [[email protected]] Sent: Wednesday, July 30, 2014 11:40 PM To: MOAB Dev Subject: [MOAB-dev] 137GB mesh file
I have an ABTR model (billions of hex elements, created by MeshKit/rgg) that is 137GB on fusion@ANL: /homes/jain/sigma/moab/lasso/meshkit/rgg/test/c_abtr_sh
I'm unable to do mbsize on it -> any suggestions? Just want to make sure if this is a valid file, also want to know material/boundary sets contained in it.
Rajeev
participants (2)
-
Grindeanu, Iulian R. -
Rajeev Jain