Smith, R. D., Dukowicz, J. K., and Malone, R. C., Parallel ocean general circulation modeling, Physica D (to appear).  Home/Search   Document Not in Database   Summary   Related Articles   Check

....places one subdomain on each processor; this provides scalability at the expense of some non nearest neighbor communication. This may be specified as follows. 11 Iocn= r c c i c j An alternative approach is to fold the octahedral mesh so as to ensure nearest neighbor communications [37]. In this approach, each processor is allocated four subdomains. This constrains scalability, but is useful when remote communication is expensive. The alternative can be implemented simply by redefining the map procedure. If the program is to execute on a cxc mesh, with processors numbered 0 to ....

Smith, R. D., Dukowicz, J. K., and Malone, R. C., Parallel ocean general circulation modeling, Physica D (to appear).

....interfaces. On the other hand, linear systems for the implicit terms are not local, since they correspond to the discretized form of elliptic equations. Solving these global systems is performed either by a direct parallelization of classical methods, like SOR or preconditioned conjuguate gradient [17] or by domain decomposition techniques [13] 14] An important point for the purpose of this work is to emphasize that ocean models are regular applications, in the sense that the volume of computations can be estimated quite precisely as a function of the grid size and the number of processors ....

Smith R.D., J.K. Dukowicz and R.C. Malone, 1992: Parallel ocean general circulation modeling. Physica D, 60, 38-61.

.... motion) which implies in that case to solve at each time step a linear system for X(t ffi t) The parallelization of ocean models is performed by domain decomposition techniques : the geographical domain is divided into subdomains, each of them being affected to a processor (e.g. [11] [3] 12] An important point for the purpose of this work is to emphasize that ocean models are regular applications, in the sense that the volume of computations can be estimated quite precisely as a function of the grid size and the number of processors (at least, it can be measured by ....

Smith R.D., J.K. Dukowicz and R.C. Malone, 1992: Parallel ocean general circulation modeling. Physica D, 60, 38-61.

....many other considerations are sometimes more important than communication cost. For example, the vertical direction is very special in atmosphere or ocean modeling, where parallelization along vertical direction is either undesirable or simply too complicated. In these cases, typically a 2D [7, 12] or even 1D [9, 10] where parallelization along domain decomposition is adopted for the 3D fields, and vertical direction are entirely local to a processor. Remapping to other decompositions[6] are often necessary to facilitate other tasks such as spectral transform [7] polar filtering [11] and ....

R.D. Smith, J.K. Dukowicz, and R.C. Malone. Parallel ocean general circulation modeling. Physica, D60, 38, 1992. See also http://gnarly.lanl.gov/Pop/Pop.html.

....difference solution to the primitive equation. On MPP, stencil updates proceed almost identically as in sequential implementation except the update order changes slightly. If the first order time stepping scheme is used, the order remains unchanged. In barotropic component of an ocean model [6, 19, 20, 24], an elliptic equation is solved using a global linear equation via a conjugate gradient (CG) method. In atmospheric data assimilation [5, 7] a correlation equation is also solved via a conjugate gradient method. The key parameters a, fi calculated through the global summation (dot product ....

....techniques. How well these techniques perform in practical codes when the variables affect dynamics Are they adequate All these require further investigations. We plan to examine the spectral transforms in the CCM atmosphere model [8, 13] and conjugate gradient solver in the POP ocean model [24]. Global summation is just an example we used in this work. High accurate arithmetics should be used in wherever reproducibility is an issue. Acknowledgment. We thank David Sarafini for pointing to us Kahan s methods, David Bailey for the double double arithmetic codes, Sherry Li for discussions ....

R. D. Smith, J. K. Dukowicz, and R. C. Malone. Parallel Ocean General Circulation Modeling. Physica, D60:38, 1992. See web page at http://www.acl.lanl.gov/climate/ models/pop. 19

....the blocks, which was difficult in our case. Research groups that implement numerical ocean models often face the need to implement Poisson solvers for problems similar to the ones described in this chapter. The ocean model developed by a group of researchers at Los Alamos National Laboratory [105] solves two dimensional Poisson problems discretized using a 9 point stencil (as opposed to a 5 point stencil used by the MIT model being developed by John Marshall, Chris Hill, and others) They chose to use sparse explicit preconditioners, in which M is given explicitly, and therefore ....

R. D. Smith, J. K. Dukowicz, and R. C. Malone. Parallel ocean general circulation modeling. Physica D, 60(1-4):38--61, 1992.

....nite di erence solution to the primitive equation. On MPP, stencil updates proceed almost identically as in sequential implementation except the update order changes slightly. If the rst order time stepping scheme is used, the order remains unchanged. In barotropic component of an ocean model [6, 19, 20, 24], an elliptic equation is solved using a global linear equation via a conjugate gradient (CG) method. In atmospheric data assimilation [5, 7] a correlation equation is also solved via a conjugate gradient method. The key parameters ; calculated through the global summation (dot product between ....

....techniques. How well these techniques perform in practical codes when the variables a ect dynamics Are they adequate All these require further investigations. We plan to examine the spectral transforms in the CCM atmosphere model [8, 13] and conjugate gradient solver in the POP ocean model [24]. Global summation is just an example we used in this work. High accurate arithmetics should be used in wherever reproducibility is an issue. Acknowledgment. We thank David Sara ni for pointing to us Kahan s methods, David Bailey for the double double arithmetic codes, Sherry Li for discussions ....

R. D. Smith, J. K. Dukowicz, and R. C. Malone. Parallel Ocean General Circulation Modeling. Physica, D60:38, 1992. See web page at http://www.acl.lanl.gov/climate/ models/pop.

....and its influence on the global climate is a grand challenge in computational science. The ocean flow dynamics and physical processes involve a broad range of spatial and temporal scales, requiring decade long integrations at fine resolutions. Recent advances in numerical ocean modeling [1, 2, 3, 4, 6, 7, 8] has greatly increased our understanding of these processes. State of art massively parallel supercomputers provide gigabytes of memory and teraflop computations necessary to run these simulations. There are a number of critical issues regarding ocean model simulations on distributed memory ....

....because it only requires a file system able to write 1D array collectively to a single file in parallel. Although this I O work is designed for MOM3 model, much of the lessons learned and strategies developed could easily be applied to other ocean models, such as the Parallel Ocean Program (POP)[6], Miami Isopycnic Ocean Model (MICOM) 7] and other grids based climate models. The remapping module and I O module are available to public upon request (MOM3 codes are available to public at GFDL web site) The modules could be easily modified to accommodating 2D domain decomposition, from the ....

R.D. Smith, J.K. Dukowicz, and R.C. Malone. Parallel ocean general circulation modeling. Physica, D60, 38, 1992.

....calculations beyond 1 4 degree are extremely costly and have only been feasible in recent years using massively parallel processing (MPP) technology. A group at Los National Laboratory has performed a 10 year integration of a 1 5 degree ocean model on the 512 processor Connection Machine 5 [3]. Using the 256 PE Cray T3D, JPL has conducted a 30 year integration of a 1 6 degree ocean model [4] It is apparent that more eddy resolving calculations with a resolution higher than 1 6 degree are needed. A one year integration of the 1 6 degree ocean model takes about 100 hours on a Cray ....

....cycle of MPP machines, the portability of the optimized ocean model and the volume rendering software will be discussed. 2. Model Description We have selected the most widely used ocean model as our base code. This ocean model is derived from the Parallel Ocean Program (POP) developed at LANL [3], which evolved from the primitive equations ocean model [5,6] developed at NOAA Geophysical Fluid Dynamics Laboratory and later known as the and model [2] or the Modular Ocean Model (MOM) 7] The ocean model used in the present study solves the 3 dimensional primitive equations using the finite ....

[Article contains additional citation context not shown here]

Smith, Dukowicz, and Malone, Parallel Ocean General Circulation Modeling, Physics D, 60,38-61, 1992.

....and its influence on the global climate is a grand challenge in computational science. The ocean flow dynamics and physical processes involve a broad range of spatial and temporal scales, requiring decade long integrations at fine resolutions. Recent advances in numerical ocean modeling [1, 2, 3, 4, 6, 7, 8] has greatly increased our understanding of these processes. State of art massively parallel supercomputers provide gigabytes of memory and teraflop computations necessary to run these simulations. There are a number of critical issues regarding ocean model simulations on distributed memory ....

....because it only requires a file system able to write 1D array collectively to a single file in parallel. Although this I O work is designed for MOM3 model, much of the lessons learned and strategies developed could easily be applied to other ocean models, such as the Parallel Ocean Program (POP)[6], Miami Isopycnic Ocean Model (MICOM) 7] and other grids based climate models. The remapping module and I O module are available to public upon request (MOM3 codes are available to public at GFDL web site) The modules could be easily modified to accommodating 2D domain decomposition, from the ....

R.D. Smith, J.K. Dukowicz, and R.C. Malone. Parallel ocean general circulation modeling. Physica, D60, 38, 1992.

....used. In summary, our machine will perform better than Hyglac on programs communicating with packets less than 1 Kbytes in size. 4 Description of the Ocean Model The Ocean General Circulation Model (OGCM) is based on the Parallel Ocean Program (POP) developed at Los Alamos National Laboratory [3]. This ocean model evolved from the Bryan Cox 3 dimensional primitive equations ocean model [4,5] developed at NOAA Geophysical Fluid Dynamics Laboratory (GFDL) and later known as the Semtner and Chervin model or the Modular Ocean Model (MOM) 6] Currently, there are hundreds of users within ....

R.D. Smith, J.K. Dukowicz, and R.C. Malone. Parallel Ocean General Circulation Modeling. Physica, 60:38--61, 1992.

....offspring as organized here, for sufficiently coarse levels, one is forced to pay the same off processor communication penalty for every point of every grid. 4 APPLICATION TO A GLOBAL OCEAN MODELING PROBLEM The original motivation for this work came from an application in global ocean modeling. In [16] an elliptic equation is solved at each time step. This equation is differenced so that the ( frequency is in the null space of the operator. The reason for this differencing is that it is required for an energy conservation relation that is deemed to be important to long time integration of ....

R. D. Smith, J. K. Dukowicz and R. C. Malone, Parallel ocean general circulation modeling, Physica D, 60(1992), pp. 38-61.

....on each processor; this provides scalability at the expense of some non nearest neighbor communication. This may be specified as follows. map(c,r,i,j,locn) fj j locn = r c c i c j g An alternative approach is to fold the octahedral mesh so as to ensure nearest neighbor communications [37]. In this approach, each processor is allocated four subdomains. This constrains scalability, but is useful when remote communication is expensive. The alternative can be implemented simply by redefining the map procedure. If the program is to execute on a c Thetac mesh, with processors numbered 0 ....

Smith, R. D., Dukowicz, J. K., and Malone, R. C., Parallel ocean general circulation modeling, Physica D (to appear).

....latitude and longitude. By not decomposing in the depth direction, we avoid having idle processors when solving the two dimensional scalar elliptic equation that arises in the barotropic section of the code, and we also avoid communication when computing the value of vertically averaged variables [5]. The code also assumes the machine topology is a mesh with a one to one mapping between cells of the tensor product grid and the processors of the machine. This assumption was made since the Intel Paragon has its processors laid out as a two dimensional mesh. Each processor only receives a local ....

....are more complicated when land mass is added to the interior of the domain. In this case, the ocean stream function may be a different constant along the boundary of each island. This can be expensive in terms of message passing since the boundary conditions for this equation are global. In [5] the barotropic equations are reformulated so that the boundary conditions are local at the expense of making the equation itself more complex. The algorithm used to solve the implicit system of equations is the conjugate gradient algorithm. As seen in [4] this involves the parallelization of ....

R. D. Smith, J. K. Dukowisz, and R. C. Malone, Parallel ocean general circulation modeling, Los Alamos National Laboratory Tech. Rep., LA-UR-92-200.

....allowing the user to choose the communication protocol and providing a relatively robust error detection mechanism so models can shut down gracefully if a particular component stops prematurely. The ocean model is the Los Alamos Parallel Ocean Program (POP) developed by Smith, Dukowicz and Malone [7] based on earlier models by Bryan [8] Cox [9] Semtner [10] and Chervin [11] The POP model was written specifically for parallel machines and supports a variety of programming models, including message passing, shared memory and data parallel. The model integrates the primitive equations using a ....

R.D. Smith, J.K. Dukowicz and R.C. Malone, Parallel ocean general circulation modeling, Physica D, 60, 38 (1992).

No context found.

Smith, Dukowicz, and Malone, Parallel Ocean General Circulation Modeling, Physics D, 60,38-61, 1992.

No context found.

Smith, and Malone, "Parallel Ocean General Circulation Modeling," Physics D, 60,38-61, 1992.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC