i'm looking to either join a group or get a group together for the final project. More precisely i would be either: interested in joining a group that uses MPI and looks at something like finite difference, finite element, solving linear sytem of equations, ... or: i have a openMP finite element code (in fortran) for dynamic rupture and elastic wave propagation which i want to parallelize using MPI since we have a little 32 node/64 processor cluster in the dept (Earth Science) on which i want to run the code. The code was for example used to show that in the case of a big earthquake on the southern San Andreas fault the San Gabriel mountains shield part of the energy going into LA. There are two versions, one using a structured mesh, and one using an unstructured mesh. The one with the unstructured mesh might be pretty hard to parallelize, the structured mesh should be ok for a final project but one would have to take a closer look at the code.
Please let me know in case you have a project in mind or would like to work on the FEM code
I am also looking for partner(s) for the project. I am open to ideas on the project, although I was thinking of some sort of study on considerations for a performant parallel code, mainly in the following 2 areas.
- Shared mem (UPC etc) vs Message Passing (MPI) - Data locality
Let me know if you are interested, and we can discuss it further. Anything interesting is fine with me.
Thanks, Manish
On 5/2/07, Jan Schmedes <ja...@umail.ucsb.edu> wrote:
> i'm looking to either join a group or get a group together for the final > project. More precisely i would be > either: > interested in joining a group that uses MPI and looks at something like > finite difference, finite element, solving linear sytem of equations, ... > or: > i have a openMP finite element code (in fortran) for dynamic rupture and > elastic wave propagation which i want to parallelize using MPI since we > have a little 32 node/64 processor cluster in the dept (Earth Science) > on which i want to run the code. The code was for example used to show > that in the case of a big earthquake on the southern San Andreas fault > the San Gabriel mountains shield part of the energy going into LA. There > are two versions, one using a structured mesh, and one using an > unstructured mesh. The one with the unstructured mesh might be pretty > hard to parallelize, the structured mesh should be ok for a final > project but one would have to take a closer look at the code.
> Please let me know in case you have a project in mind or would like to > work on the FEM code
> I am also looking for partner(s) for the project. > I am open to ideas on the project, although I was thinking of some sort of > study on considerations for a performant parallel code, mainly in the > following 2 areas.
> - Shared mem (UPC etc) vs Message Passing (MPI) > - Data locality
> Let me know if you are interested, and we can discuss it further. Anything > interesting is fine with me.
> Thanks, > Manish
> On 5/2/07, Jan Schmedes <j...@umail.ucsb.edu> wrote:
> > Hi all,
> > i'm looking to either join a group or get a group together for the final > > project. More precisely i would be > > either: > > interested in joining a group that uses MPI and looks at something like > > finite difference, finite element, solving linear sytem of equations, ... > > or: > > i have a openMP finite element code (in fortran) for dynamic rupture and > > elastic wave propagation which i want to parallelize using MPI since we > > have a little 32 node/64 processor cluster in the dept (Earth Science) > > on which i want to run the code. The code was for example used to show > > that in the case of a big earthquake on the southern San Andreas fault > > the San Gabriel mountains shield part of the energy going into LA. There > > are two versions, one using a structured mesh, and one using an > > unstructured mesh. The one with the unstructured mesh might be pretty > > hard to parallelize, the structured mesh should be ok for a final > > project but one would have to take a closer look at the code.
> > Please let me know in case you have a project in mind or would like to > > work on the FEM code
I certainly would prefer MPI too. Now two my second suggestion in my last email: i want to parallelize this code anyway for my research so from my point of view this is certainly a good project. But as i mentioned, it is written in fortran (i would prefer C but i did not write the code...). Now to Manish suggestion for a comparison. The code is already in openMP and we should be able to run it on a node on datastar. If we parallelize it we could make a comparison between openMP and MPI. But since i did not look too much at the code yet i don't know how much work it would be to parallelize it since this also involves some understanding of what the code does. But since i want to do that anyway i don't mind spending 'extra' time on that... Very interesting would of course the code with the unstructured mesh, but i think this will be too hard given the limited time frame unless there are tools available that can help partitioning the mesh (anyone knows something about that topic?).
Andreas Bach wrote: > I need a group/parner as well.
> I'm from CS. Haven't decided any preference of topic, but I prefer MPI > over UPC (:
> Let me know...
> On May 2, 8:26 pm, "Manish Goyal" <bhu...@gmail.com> wrote:
>> Jan,
>> I am also looking for partner(s) for the project. >> I am open to ideas on the project, although I was thinking of some sort of >> study on considerations for a performant parallel code, mainly in the >> following 2 areas.
>> - Shared mem (UPC etc) vs Message Passing (MPI) >> - Data locality
>> Let me know if you are interested, and we can discuss it further. Anything >> interesting is fine with me.
>> Thanks, >> Manish
>> On 5/2/07, Jan Schmedes <j...@umail.ucsb.edu> wrote:
>>> Hi all,
>>> i'm looking to either join a group or get a group together for the final >>> project. More precisely i would be >>> either: >>> interested in joining a group that uses MPI and looks at something like >>> finite difference, finite element, solving linear sytem of equations, ... >>> or: >>> i have a openMP finite element code (in fortran) for dynamic rupture and >>> elastic wave propagation which i want to parallelize using MPI since we >>> have a little 32 node/64 processor cluster in the dept (Earth Science) >>> on which i want to run the code. The code was for example used to show >>> that in the case of a big earthquake on the southern San Andreas fault >>> the San Gabriel mountains shield part of the energy going into LA. There >>> are two versions, one using a structured mesh, and one using an >>> unstructured mesh. The one with the unstructured mesh might be pretty >>> hard to parallelize, the structured mesh should be ok for a final >>> project but one would have to take a closer look at the code.
>>> Please let me know in case you have a project in mind or would like to >>> work on the FEM code
Being from computer science, it will be interesting to work on something from a different area. I donot know much about the algorithm that Jan is talking about, but surely we can take a look at it to see what can be done in the time period, and if we could extra, that would be great.
I donot mind MPI, would rather prefer MPI as it is clean and would need less efforts and focus, and we can spend more time on the problem on hand.
Manish On 5/3/07, Jan Schmedes <ja...@umail.ucsb.edu> wrote:
> I certainly would prefer MPI too. Now two my second suggestion in my > last email: > i want to parallelize this code anyway for my research so from my point > of view this is certainly a good project. But as i mentioned, it is > written in fortran (i would prefer C but i did not write the code...). > Now to Manish suggestion for a comparison. The code is already in openMP > and we should be able to run it on a node on datastar. If we parallelize > it we could make a comparison between openMP and MPI. But since i did > not look too much at the code yet i don't know how much work it would > be to parallelize it since this also involves some understanding of what > the code does. But since i want to do that anyway i don't mind spending > 'extra' time on that... > Very interesting would of course the code with the unstructured mesh, > but i think this will be too hard given the limited time frame unless > there are tools available that can help partitioning the mesh (anyone > knows something about that topic?).
> Please let me know what your thoughts are
> Cheers
> Jan
> Andreas Bach wrote: > > I need a group/parner as well.
> > I'm from CS. Haven't decided any preference of topic, but I prefer MPI > > over UPC (:
> > Let me know...
> > On May 2, 8:26 pm, "Manish Goyal" <bhu...@gmail.com> wrote:
> >> Jan,
> >> I am also looking for partner(s) for the project. > >> I am open to ideas on the project, although I was thinking of some sort > of > >> study on considerations for a performant parallel code, mainly in the > >> following 2 areas.
> >> - Shared mem (UPC etc) vs Message Passing (MPI) > >> - Data locality
> >> Let me know if you are interested, and we can discuss it further. > Anything > >> interesting is fine with me.
> >> Thanks, > >> Manish
> >> On 5/2/07, Jan Schmedes <j...@umail.ucsb.edu> wrote:
> >>> Hi all,
> >>> i'm looking to either join a group or get a group together for the > final > >>> project. More precisely i would be > >>> either: > >>> interested in joining a group that uses MPI and looks at something > like > >>> finite difference, finite element, solving linear sytem of equations, > ... > >>> or: > >>> i have a openMP finite element code (in fortran) for dynamic rupture > and > >>> elastic wave propagation which i want to parallelize using MPI since > we > >>> have a little 32 node/64 processor cluster in the dept (Earth Science) > >>> on which i want to run the code. The code was for example used to > show > >>> that in the case of a big earthquake on the southern San Andreas fault > >>> the San Gabriel mountains shield part of the energy going into LA. > There > >>> are two versions, one using a structured mesh, and one using an > >>> unstructured mesh. The one with the unstructured mesh might be pretty > >>> hard to parallelize, the structured mesh should be ok for a final > >>> project but one would have to take a closer look at the code.
> >>> Please let me know in case you have a project in mind or would like to > >>> work on the FEM code
Sounds good, and fortran is ok for you? What about looking at the code and have a little discussion for example Monday after class? I will have a closer look at the code before that.
Manish Goyal wrote: > Being from computer science, it will be interesting to work on > something from a different area. I donot know much about the algorithm > that Jan is talking about, but surely we can take a look at it to see > what can be done in the time period, and if we could extra, that would > be great.
> I donot mind MPI, would rather prefer MPI as it is clean and would > need less efforts and focus, and we can spend more time on the > problem on hand.
> I certainly would prefer MPI too. Now two my second suggestion in my > last email: > i want to parallelize this code anyway for my research so from my > point > of view this is certainly a good project. But as i mentioned, it is > written in fortran (i would prefer C but i did not write the code...). > Now to Manish suggestion for a comparison. The code is already in > openMP > and we should be able to run it on a node on datastar. If we > parallelize > it we could make a comparison between openMP and MPI. But since i did > not look too much at the code yet i don't know how much work it > would > be to parallelize it since this also involves some understanding > of what > the code does. But since i want to do that anyway i don't mind > spending > 'extra' time on that... > Very interesting would of course the code with the unstructured mesh, > but i think this will be too hard given the limited time frame unless > there are tools available that can help partitioning the mesh (anyone > knows something about that topic?).
> Please let me know what your thoughts are
> Cheers
> Jan
> Andreas Bach wrote: > > I need a group/parner as well.
> > I'm from CS. Haven't decided any preference of topic, but I > prefer MPI > > over UPC (:
> > Let me know...
> > On May 2, 8:26 pm, "Manish Goyal" <bhu...@gmail.com > <mailto:bhu...@gmail.com>> wrote:
> >> Jan,
> >> I am also looking for partner(s) for the project. > >> I am open to ideas on the project, although I was thinking of > some sort of > >> study on considerations for a performant parallel code, mainly > in the > >> following 2 areas.
> >> - Shared mem (UPC etc) vs Message Passing (MPI) > >> - Data locality
> >> Let me know if you are interested, and we can discuss it > further. Anything > >> interesting is fine with me.
> >> Thanks, > >> Manish
> >> On 5/2/07, Jan Schmedes <j...@umail.ucsb.edu > <mailto:j...@umail.ucsb.edu>> wrote:
> >>> Hi all,
> >>> i'm looking to either join a group or get a group together for > the final > >>> project. More precisely i would be > >>> either: > >>> interested in joining a group that uses MPI and looks at > something like > >>> finite difference, finite element, solving linear sytem of > equations, ... > >>> or: > >>> i have a openMP finite element code (in fortran) for dynamic > rupture and > >>> elastic wave propagation which i want to parallelize using MPI > since we > >>> have a little 32 node/64 processor cluster in the dept (Earth > Science) > >>> on which i want to run the code. The code was for example > used to show > >>> that in the case of a big earthquake on the southern San > Andreas fault > >>> the San Gabriel mountains shield part of the energy going into > LA. There > >>> are two versions, one using a structured mesh, and one using an > >>> unstructured mesh. The one with the unstructured mesh might be > pretty > >>> hard to parallelize, the structured mesh should be ok for a final > >>> project but one would have to take a closer look at the code.
> >>> Please let me know in case you have a project in mind or would > like to > >>> work on the FEM code
> Sounds good, and fortran is ok for you? What about looking at the code > and have a little discussion for example Monday after class? I will have > a closer look at the code before that.
> Cheers
> Jan
> Manish Goyal wrote: > > Being from computer science, it will be interesting to work on > > something from a different area. I donot know much about the algorithm > > that Jan is talking about, but surely we can take a look at it to see > > what can be done in the time period, and if we could extra, that would > > be great.
> > I donot mind MPI, would rather prefer MPI as it is clean and would > > need less efforts and focus, and we can spend more time on the > > problem on hand.
> > I certainly would prefer MPI too. Now two my second suggestion in my > > last email: > > i want to parallelize this code anyway for my research so from my > > point > > of view this is certainly a good project. But as i mentioned, it is > > written in fortran (i would prefer C but i did not write the code...). > > Now to Manish suggestion for a comparison. The code is already in > > openMP > > and we should be able to run it on a node on datastar. If we > > parallelize > > it we could make a comparison between openMP and MPI. But since i did > > not look too much at the code yet i don't know how much work it > > would > > be to parallelize it since this also involves some understanding > > of what > > the code does. But since i want to do that anyway i don't mind > > spending > > 'extra' time on that... > > Very interesting would of course the code with the unstructured mesh, > > but i think this will be too hard given the limited time frame unless > > there are tools available that can help partitioning the mesh (anyone > > knows something about that topic?).
> > Please let me know what your thoughts are
> > Cheers
> > Jan
> > Andreas Bach wrote: > > > I need a group/parner as well.
> > > I'm from CS. Haven't decided any preference of topic, but I > > prefer MPI > > > over UPC (:
> > > Let me know...
> > > On May 2, 8:26 pm, "Manish Goyal" <bhu...@gmail.com > > <mailto:bhu...@gmail.com>> wrote:
> > >> Jan,
> > >> I am also looking for partner(s) for the project. > > >> I am open to ideas on the project, although I was thinking of > > some sort of > > >> study on considerations for a performant parallel code, mainly > > in the > > >> following 2 areas.
> > >> - Shared mem (UPC etc) vs Message Passing (MPI) > > >> - Data locality
> > >> Let me know if you are interested, and we can discuss it > > further. Anything > > >> interesting is fine with me.
> > >> Thanks, > > >> Manish
> > >> On 5/2/07, Jan Schmedes <j...@umail.ucsb.edu > > <mailto:j...@umail.ucsb.edu>> wrote:
> > >>> Hi all,
> > >>> i'm looking to either join a group or get a group together for > > the final > > >>> project. More precisely i would be > > >>> either: > > >>> interested in joining a group that uses MPI and looks at > > something like > > >>> finite difference, finite element, solving linear sytem of > > equations, ... > > >>> or: > > >>> i have a openMP finite element code (in fortran) for dynamic > > rupture and > > >>> elastic wave propagation which i want to parallelize using MPI > > since we > > >>> have a little 32 node/64 processor cluster in the dept (Earth > > Science) > > >>> on which i want to run the code. The code was for example > > used to show > > >>> that in the case of a big earthquake on the southern San > > Andreas fault > > >>> the San Gabriel mountains shield part of the energy going into > > LA. There > > >>> are two versions, one using a structured mesh, and one using an > > >>> unstructured mesh. The one with the unstructured mesh might be > > pretty > > >>> hard to parallelize, the structured mesh should be ok for a final > > >>> project but one would have to take a closer look at the code.
> > >>> Please let me know in case you have a project in mind or would > > like to > > >>> work on the FEM code
Andreas Bach wrote: > Can I join in with you guys?
> On May 4, 9:37 am, Jan Schmedes <j...@umail.ucsb.edu> wrote:
>> Sounds good, and fortran is ok for you? What about looking at the code >> and have a little discussion for example Monday after class? I will have >> a closer look at the code before that.
>> Cheers
>> Jan
>> Manish Goyal wrote:
>>> Being from computer science, it will be interesting to work on >>> something from a different area. I donot know much about the algorithm >>> that Jan is talking about, but surely we can take a look at it to see >>> what can be done in the time period, and if we could extra, that would >>> be great.
>>> I donot mind MPI, would rather prefer MPI as it is clean and would >>> need less efforts and focus, and we can spend more time on the >>> problem on hand.
>>> I certainly would prefer MPI too. Now two my second suggestion in my >>> last email: >>> i want to parallelize this code anyway for my research so from my >>> point >>> of view this is certainly a good project. But as i mentioned, it is >>> written in fortran (i would prefer C but i did not write the code...). >>> Now to Manish suggestion for a comparison. The code is already in >>> openMP >>> and we should be able to run it on a node on datastar. If we >>> parallelize >>> it we could make a comparison between openMP and MPI. But since i did >>> not look too much at the code yet i don't know how much work it >>> would >>> be to parallelize it since this also involves some understanding >>> of what >>> the code does. But since i want to do that anyway i don't mind >>> spending >>> 'extra' time on that... >>> Very interesting would of course the code with the unstructured mesh, >>> but i think this will be too hard given the limited time frame unless >>> there are tools available that can help partitioning the mesh (anyone >>> knows something about that topic?).
>>> Please let me know what your thoughts are
>>> Cheers
>>> Jan
>>> Andreas Bach wrote: >>> > I need a group/parner as well.
>>> > I'm from CS. Haven't decided any preference of topic, but I >>> prefer MPI >>> > over UPC (:
>>> > Let me know...
>>> > On May 2, 8:26 pm, "Manish Goyal" <bhu...@gmail.com >>> <mailto:bhu...@gmail.com>> wrote:
>>> >> Jan,
>>> >> I am also looking for partner(s) for the project. >>> >> I am open to ideas on the project, although I was thinking of >>> some sort of >>> >> study on considerations for a performant parallel code, mainly >>> in the >>> >> following 2 areas.
>>> >> - Shared mem (UPC etc) vs Message Passing (MPI) >>> >> - Data locality
>>> >> Let me know if you are interested, and we can discuss it >>> further. Anything >>> >> interesting is fine with me.
>>> >> Thanks, >>> >> Manish
>>> >> On 5/2/07, Jan Schmedes <j...@umail.ucsb.edu >>> <mailto:j...@umail.ucsb.edu>> wrote:
>>> >>> Hi all,
>>> >>> i'm looking to either join a group or get a group together for >>> the final >>> >>> project. More precisely i would be >>> >>> either: >>> >>> interested in joining a group that uses MPI and looks at >>> something like >>> >>> finite difference, finite element, solving linear sytem of >>> equations, ... >>> >>> or: >>> >>> i have a openMP finite element code (in fortran) for dynamic >>> rupture and >>> >>> elastic wave propagation which i want to parallelize using MPI >>> since we >>> >>> have a little 32 node/64 processor cluster in the dept (Earth >>> Science) >>> >>> on which i want to run the code. The code was for example >>> used to show >>> >>> that in the case of a big earthquake on the southern San >>> Andreas fault >>> >>> the San Gabriel mountains shield part of the energy going into >>> LA. There >>> >>> are two versions, one using a structured mesh, and one using an >>> >>> unstructured mesh. The one with the unstructured mesh might be >>> pretty >>> >>> hard to parallelize, the structured mesh should be ok for a final >>> >>> project but one would have to take a closer look at the code.
>>> >>> Please let me know in case you have a project in mind or would >>> like to >>> >>> work on the FEM code
> Andreas Bach wrote: > > Can I join in with you guys?
> > On May 4, 9:37 am, Jan Schmedes <j...@umail.ucsb.edu> wrote:
> >> Sounds good, and fortran is ok for you? What about looking at the code > >> and have a little discussion for example Monday after class? I will have > >> a closer look at the code before that.
> >> Cheers
> >> Jan
> >> Manish Goyal wrote:
> >>> Being from computer science, it will be interesting to work on > >>> something from a different area. I donot know much about the algorithm > >>> that Jan is talking about, but surely we can take a look at it to see > >>> what can be done in the time period, and if we could extra, that would > >>> be great.
> >>> I donot mind MPI, would rather prefer MPI as it is clean and would > >>> need less efforts and focus, and we can spend more time on the > >>> problem on hand.
> >>> I certainly would prefer MPI too. Now two my second suggestion in my > >>> last email: > >>> i want to parallelize this code anyway for my research so from my > >>> point > >>> of view this is certainly a good project. But as i mentioned, it is > >>> written in fortran (i would prefer C but i did not write the code...). > >>> Now to Manish suggestion for a comparison. The code is already in > >>> openMP > >>> and we should be able to run it on a node on datastar. If we > >>> parallelize > >>> it we could make a comparison between openMP and MPI. But since i did > >>> not look too much at the code yet i don't know how much work it > >>> would > >>> be to parallelize it since this also involves some understanding > >>> of what > >>> the code does. But since i want to do that anyway i don't mind > >>> spending > >>> 'extra' time on that... > >>> Very interesting would of course the code with the unstructured mesh, > >>> but i think this will be too hard given the limited time frame unless > >>> there are tools available that can help partitioning the mesh (anyone > >>> knows something about that topic?).
> >>> Please let me know what your thoughts are
> >>> Cheers
> >>> Jan
> >>> Andreas Bach wrote: > >>> > I need a group/parner as well.
> >>> > I'm from CS. Haven't decided any preference of topic, but I > >>> prefer MPI > >>> > over UPC (:
> >>> > Let me know...
> >>> > On May 2, 8:26 pm, "Manish Goyal" <bhu...@gmail.com > >>> <mailto:bhu...@gmail.com>> wrote:
> >>> >> Jan,
> >>> >> I am also looking for partner(s) for the project. > >>> >> I am open to ideas on the project, although I was thinking of > >>> some sort of > >>> >> study on considerations for a performant parallel code, mainly > >>> in the > >>> >> following 2 areas.
> >>> >> - Shared mem (UPC etc) vs Message Passing (MPI) > >>> >> - Data locality
> >>> >> Let me know if you are interested, and we can discuss it > >>> further. Anything > >>> >> interesting is fine with me.
> >>> >> Thanks, > >>> >> Manish
> >>> >> On 5/2/07, Jan Schmedes <j...@umail.ucsb.edu > >>> <mailto:j...@umail.ucsb.edu>> wrote:
> >>> >>> Hi all,
> >>> >>> i'm looking to either join a group or get a group together for > >>> the final > >>> >>> project. More precisely i would be > >>> >>> either: > >>> >>> interested in joining a group that uses MPI and looks at > >>> something like > >>> >>> finite difference, finite element, solving linear sytem of > >>> equations, ... > >>> >>> or: > >>> >>> i have a openMP finite element code (in fortran) for dynamic > >>> rupture and > >>> >>> elastic wave propagation which i want to parallelize using MPI > >>> since we > >>> >>> have a little 32 node/64 processor cluster in the dept (Earth > >>> Science) > >>> >>> on which i want to run the code. The code was for example > >>> used to show > >>> >>> that in the case of a big earthquake on the southern San > >>> Andreas fault > >>> >>> the San Gabriel mountains shield part of the energy going into > >>> LA. There > >>> >>> are two versions, one using a structured mesh, and one using an > >>> >>> unstructured mesh. The one with the unstructured mesh might be > >>> pretty > >>> >>> hard to parallelize, the structured mesh should be ok for a final > >>> >>> project but one would have to take a closer look at the code.
> >>> >>> Please let me know in case you have a project in mind or would > >>> like to > >>> >>> work on the FEM code
