A. Choudhary, G. Fox, S. Ranka, S. Himnandani, K. Kennedy, C. Koelbel, and J. Saltz. Software Support for Irregular and Loosely Synchronous Problems. International Journal of Computing Systems in Engineering, 3 (1-4), 1992.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....making use of a vendor s library can improve performance. 3. Dynamic Redistribution of Data The technique of dynamically redistributing data such that each processor has a uniform workload is an essential operation in many irregular problems, such as computational adaptive graph (grid)problems ([27, 16, 12]) including finite element calculations, molecular dynamics [21] particle dy namics [15] plasma particle4n cell [17] raytraced volume rendering [19] region growing and computer vision [30] and statistical physics [8] Here, the input is distributed across p processors with a distribution ....

A. Choudhary, G. Fox, S. Ranka, S. Himnandani, K. Kennedy, C. Koelbel, and J. Saltz. Software Support for Irregular and Loosely Synchronous Problems. International Journal of Computing Systems in Engineering, 3 (1-4), 1992.

....efficient parallelizations of dynamic and irregular problems have received much attention only recently. Many high level languages like Fortran D [3] ViennaFortran [12] support parallel operations on uniform arrays. Some runtime support for irregular structures is however available in PARTI [7], but these approaches take advantage of static data access and communication patterns. These do not provide efficient solutions for dynamically changing data distribution and communication patterns as are seen in the problems we are addressing here. On parallel machines available today the cost ....

Choudhary, A., Fox, G., Hiranandani, S., Kennedy, K., Koelbel, C., Ranka, S., Saltz, J., Software Support for Irregular and Loosely Synchronous Problems, Proceedings of the Conference on High Performance Computing for Flight Vehicles, 1992.

....are to execute concurrently, for establishing typed, one to one communication channels between processes, and for sending and receiving messages on channels. Fortran M is built on the paradigm of task parallelism, and can be used to coordinate multiple data parallel computations. In recent work [12], the authors Choudhury et al. assert that parallel computing demands support from software that precisely and effectively captures the structure of the application for better performance. They define about ten broad classes of computations, each of which is large enough to warrant individually ....

....The trend is toward developing automatic run time support software and linking it to compilers, in order to parallelize large real world unstructured sparse matrix problems. Increasing emphasis is being placed on having the compiler itself generate parts of the run time software. In [12], the authors conclude that for the implicit multiphase loosely synchronous problems, there is still a clear need for development of appropriate run time support targeted toward SIMD and MIMD distributed memory architectures. Our goal is to start building such a system for large scale ....

A. Choudhary, G. Fox, S. Hiranandani, K. Kennedy, C. Koelbel, S. Ranka, and J. Saltz. Software support for irregular and loosely synchronous problems. Computing Systems in Engineering, 3(1):43--52, 1992.

....This library would assume the r oles of the directives and language extensions in HPF as well as the HPF library. We will loosely distinguish two di erent levels at which a library implementation of the HPF semantics can operate. The rst is the level of the so called run time libraries [1, 8, 9, 6]. This kind of library provides functions for scheduling and executing speci c patterns of collective communication already identi ed by a compiler (in the HPF case) or else by an application programmer using the library directly. Such a library may also provide functions for translating between ....

A. Choudhary, G. Fox, S. Ranka, S. Hiranandani, K. Kennedy, C. Koelbel, and J. Saltz. Software support for irregular and loosely synchronous problems. Computing Systems in Engineering, 3:43-52, 1992.

....directives and language extensions in HPF as well as the HPF library. We will loosely distinguish two different levels at which a library implementation of the HPF semantics (or, at least, the HPF distributed data model) can operate. The first is the level of the so called run time libraries [1, 7, 8, 5]. This kind of library provides functions for scheduling and executing specific patterns of collective communication already identified by a compiler (in the HPF case) or else by an application programmer using the library directly. Such a library may also provide functions for translating between ....

A. Choudhary, G. Fox, S. Ranka, S. Hiranandani, K. Kennedy, C. Koelbel, and J. Saltz. Software support for irregular and loosely synchronous problems. Computing Systems in Engineering, 3:43--52, 1992.

....correct one dynamically at run time in a preprocessing step that takes place right after the parameters in question are known. This strategy is similar to the dynamic compilation strategies in Java [19] and the inspector executor method for compiling irregular applications for parallel execution [5]. Although these strategies are not unique to telescoping languages, the framework makes it possible to automate many of the steps in tailoring libraries so that the library developer need only identify shared compute intensive kernels along with test drivers leaving the specific optimizations ....

A. Choudhary, G. Fox, S. Ranka, S. Hiranandani, K. Kennedy, C. Koelbel, and J. Saltz. Software support for irregular and loosely synchronous problems. International Journal of Computing Systems in Engineering, 3(2):43--52, 1993.

.... after third and final merge iteration Figure 2: The Merge Stage Parallel Implementations The region growing problem is a representative of a type of loosely synchronous problems, known as adaptive irregular problems, whose data objects evolve during the computation in a time synchronized manner [4]. The problem exhibits a dynamic behavior that starts with a high degree of parallelism that very rapidly diminishes to a much lower degree of parallelism. The split and merge algorithm for solving the region growing problem was implemented in both the data parallel and message passing models. In ....

G. Fox et al, "Software support for irregular and loosely synchronous problems", Technical Report, May 1992, Northeast Parallel Architectures Center, Syracuse University.

....or three dimensional coordinates, and the interaction between computations is limited to vertices that are physically proximate. Examples of such applications include finite element calculations, molecular dynamics, particle dynamics, particle in a cell, region growing, and statistical physics [3]. For these applications, partitioning Northeast Parallel Architectures Center and School of Computer and Information Science, Syracuse University, Syracuse, NY 13244 4100. This research was supported in part by DARPA under contract #DABT63 91 C 0028. can be achieved by exploiting the above ....

....recursive coordinate bisection [19] inertial bisection [6] etc. We have discussed an index based indexing scheme in [15] and shown that it produces good mappings for computational structures satisfying the above property. For a large class of irregular and adaptive data parallel applications [3], the computational structure changes from one phase to another in an incremental fashion. The following scenarios may arise: ffl Perturbation: All the coordinates may perturb (within some small distance) e.g. particledynamics problems [14] ffl Vertex Additions: New vertices may be added ....

[Article contains additional citation context not shown here]

A. Choudhary, G. Fox, S. Hiranandani, K. Kennedy, C. Koelbel, S. Ranka, and J. Saltz. Software Support for Irregular and Loosely Synchronous Problems. Proceedings of the Conference on High Performance Computing for Flight Vehicles, 1992.

....and pass them as parameters to NICE along with the CPT. The structure of these message buffers is given by a nested pointer structure as the one shown in Figure 9. 5 Using NICE Primitives in Unstructured Applications In this section, we present a simple example (from examples discussed in [4]) to demonstrate the use of NICE primitives. A static single phase computation consists of a single concurrent computational phase, which may be executed repeatedly without change [4] Examples of static single phase computations, which are iterative solvers using sparse matrix vector ....

....Primitives in Unstructured Applications In this section, we present a simple example (from examples discussed in [4] to demonstrate the use of NICE primitives. A static single phase computation consists of a single concurrent computational phase, which may be executed repeatedly without change [4]. Examples of static single phase computations, which are iterative solvers using sparse matrix vector multiplications, can be found in [13] Examples of explicit unstructured mesh fluids calculations can be found in [14] Figure 10 depicts a schematic outline of a kernel from a fluid dynamics ....

Alok Choudhary, Geoffrey C. Fox, Seema Hiranandani, Ken Kennedy, Charles Koelbel, Sanjay Ranka, and Joel Saltz. Software support for irregular and loosely synchronous problems. In Proceedings of the Conference on High Performance Computing for Flight Vehicles, 1992. To appear.

....Ties are Broken by Choosing the Neighbor With the Smallest ID 2. 3 Resolving Ties at Random The region growing problem is a representative of a type of loosely synchronous problems, known as adaptive irregular problems, whose data objects evolve during the computation in a time synchronized manner [4]. The problem exhibits a dynamic behavior that starts with a high degree of parallelism that very rapidly diminishes to a much lower degree of parallelism. In order to increase the degree of parallelism in the algorithm, we introduced an element of randomness to our parallel implementations. In ....

G. Fox et al, "Software support for irregular and loosely synchronous problems", Technical Report, Northeast Parallel Architectures Center, Syracuse University, May 1992.

....and optimizations for compiling irregular applications. Specifically, we concentrate on applications in which data is accessed using indirection arrays. Such codes are common in computational fluid dynamics, molecular dynamics, in Particle In Cell (PIC) problems and in numerical simulations [10]. The commonly used approach for compiling irregular applications is the inspector executor model [27] Conceptually, an inspector or a communication preprocessing statement analyses the indirection array to determine the communication required by a data parallel loop. The results of communication ....

A. Choudhary, G. Fox, S. Hiranandani, K. Kennedy, C. Koelbel, S. Ranka, and J. Saltz. Software support for irregular and loosely synchronous problems. Computing Systems in Engineering, 3(1--4):43--52, 1992. Papers presented at the Symposium on High-Performance Computing for Flight Vehicles, December 1992.

....subdomains. Examples include multiblock Navier Stokes solvers and structured adaptive multigrid problems. We will call this class of problems irregularly coupled regular mesh computations (ICRMs) In a different paper in this volume, a more detailed taxonomy of irregular problems is presented [11]. In the kinds of algorithms we consider here, data produced or input during a program s initialization phase play a large role in determining the nature of the subsequent 1 This work was supported by NASA contract NAS1 18605 while the authors were in residence at ICASE, NASA Langley Research ....

Alok Choudhary, Geoffrey Fox, Sanjay Ranka, Seema Hiranandani, Ken Kennedy, Charles Koelbel, and Joel Saltz. Software support for irregular and loosely synchronous problems. In Proceedings of the Symposium on High-Performance Computing for Flight Vehicles, December 1992.

....is limited to vertices that are physically proximate. Examples of such applications include finite element calculations [21] molecular dynamics [3] particle dynamics [25] particle ina cell [17, 31] region growing [6] and statistical physics [5] A list of other such applications is given in [4]. For these applications, partitioning can be achieved by exploiting the above property. Essentially proximate points are clustered together and form a partition such that the number of points attached to each partition are approximately equal. Most of the interactions are local and the amount of ....

....recursive coordinate bisection [32] and inertial bisection [14] We have discussed an index based indexing scheme in [27] and shown that it produces good mappings for computational structures satisfying the above property. For a large class of irregular and adaptive data parallel applications [4], the computational structure changes from one phase to another in an incremental fashion. Thus the partitioning information of the previous phase can be effectively utilized to give the partitioning for a new phase. Changes are typically gradual, reflecting adiabatic changes in the physical ....

[Article contains additional citation context not shown here]

A. Choudhary, G. Fox, S. Hiranandani, K. Kennedy, C. Koelbel, S. Ranka, and J. Saltz. Software Support for Irregular and Loosely Synchronous Problems. Proceedings of the Conference on High Performance Computing for Flight Vehicles, 1992.

....distributed memory parallel computers. It is important to map the program such that the total execution time is minimized; the mapping can typically be performed statically or dynamically. For a large class of scientific problems that are irregular in nature, achieving a good mapping is difficult [1]. The nature of the irregularities is unknown at the time of compilation and can be derived only at runtime. The handling of such irregular problems requires runtime information in order to partition the computation in such a fashion that each processor receives an approximately equal amount of ....

....optimization methods are gradient descent methods and hence require minimization of a differentiable function. 2. 1 Incremental problems An adaptive irregular computation consists of a loosely synchronous computation executed repeatedly in which the data access pattern changes between iterations [1]. The changes may be gradual, reflecting adiabatic changes in the physical domain (e.g. molecular dynamics) or large scale reflecting additions to a data structure (e.g. adaptive PDE solvers) The physical and numerical properties of these algorithms typically guarantee that large scale ....

Alok Choudhary, Geoffrey C. Fox, Seema Hiranandani, Ken Kennedy, Charles Koelbel, Sanjay Ranka, and Joel Saltz. Software support for irregular and loosely synchronous problems. In Proceedings of the Conference on High Performance Computing for Flight Vehicles, 1992. to appear.

....runtime support for many classes of applications. One class of scientific and engineering applications involves structured meshes. These meshes may be nested (as in multigrid or adaptive codes) or may be irregularly coupled (called Multiblock Problems or Irregularly Coupled Regular Meshes) [1, 4, 5, 14]. Multigrid codes typically have a number of meshes at different levels of resolution, so may require array section moves with non unit strides. Multigrid codes are often used in computational fluid dynamics applications. Examples of multiblock problems [5] include multiblock Navier Stokes solvers ....

....Coupled Regular Meshes) 1, 4, 5, 14] Multigrid codes typically have a number of meshes at different levels of resolution, so may require array section moves with non unit strides. Multigrid codes are often used in computational fluid dynamics applications. Examples of multiblock problems [5] include multiblock Navier Stokes solvers and structured adaptive multigrid problems [16] These problems have the following characteristics: 1 This work was supported by ARPA under contract No. NAG 1 1485, by NSF under grant No. ASC 9213821 and by ONR under contract No. SC 292 1 22913. The ....

A. Choudhary, G. Fox, S. Hiranandani, K. Kennedy, C. Koelbel, S. Ranka, and J. Saltz. Software support for irregular and loosely synchronous problems. Computing Systems in Engineering, 3(1-4):43--52, 1992. Papers presented at the Symposium on HighPerformance Computing for Flight Vehicles, December 1992.

....on the owner computes rule) 5] This typically results in regular collective communication between processors. Many such primitives have been developed in [1, 13] For a large class of scientific problems, which are irregular in nature, achieving a good mapping is considerably more difficult [6]. The nature of this irregularity may not be known at the time of compilation, and can be derived only at run time. Packages like This work was supported in part by NSF under CCR9110812 and in part by DARPA under contract #DABT6391 C 0028. The contents do not necessarily reflect the position ....

Alok Choudhary, Geoffrey C. Fox, Seema Hiranandani, Ken Kennedy, Charles Koelbel, Sanjay Ranka, and Joel Saltz. Software Support for Irregular and Loosely Synchronous Problems. Journal of Computing Systems in Engineering, 3:pp. 43--52, 1993.

....on the owner computes rule) 6] This typically results in regular collective communication between processors. Many such primitives have been developed in [1, 16] For a large class of scientific problems, which are irregular in nature, achieving a good mapping is considerably more difficult [7]. The nature of this irregularity may not be known at the time of compilation, and can be derived only at run time. Packages like PARTI [9, 12, 15] derive the necessary communication information based on the data required for performing the local computations and data partitioning. This tends to ....

Alok Choudhary, Geoffrey C. Fox, Seema Hiranandani, Ken Kennedy, Charles Koelbel, Sanjay Ranka, and Joel Saltz. Software support for irregular and loosely synchronous problems. In Proceedings of the Conference on High Performance Computing for Flight Vehicles, 1992. to appear.

No context found.

Choudhary, A., Fox, G., Ranka, S., Hiranandani, S., Kennedy, K., Koelbel, C., and Saltz, J. "Software support for irregular and loosely synchronous problems," Computing Systems in Engineering, 3(1--4):43--52, 1992. CSE-MS 118, CRPC-TR92258.

....of data at runtime, which is particularly important for dynamic and irregular problems. The user can either specify the distribution, or specify the computational graph, in which case the compiler will find a good distribution using various optimization techniques such as those described above [5, 51]. Using the physical analogy we introduced in Section 4.4, we can think of the operating system as a heat bath that keeps the computation cool and therefore near its ground state (optimal solution) Most scientific simulations change slowly with time and redistribution of processes by the ....

Choudhary, A., Fox, G., Ranka, S., Hiranandani, S., Kennedy, K., Koelbel, C. & Saltz, J. (1992). Software Support for Irregular and Loosely Synchronous Problems. In Computing Systems in Engineering Vol. 3, pp. 43--52. Pergamon Press, Oxford.

.... of compilation by giving directives to decompose the data and its corresponding computations [8] For irregular applications, achieving a good mapping is considerably more difficult; the nature of the irregularities may not be known at the time of compilation and can be derived only at runtime [7]. These applications can be represented as computational graphs from the perspective of parallel computing. The vertices of these graphs represent tasks that can be executed concurrently, while the edges represent the interactions between them. The key problem in efficiently executing irregular ....

....or three dimensional coordinates, and the interaction between computations is limited to physically proximate vertices. Examples of such applications include molecular dynamics, static and adaptive PDE solvers [30, 38] region growing, component labeling [10] and statistical physics simulations [6, 7, 9, 10]. Simple and fast heuristics for partitioning such graphs is to cluster physically proximate points in two or three dimensions. Most of the above applications are iterative and the same computational graph is used for several iterations. The average time required to solve one iteration of these ....

A. Choudhary, G. Fox, S. Hiranandani, K. Kennedy, C. Koelbel, S. Ranka, and J. Saltz. Software Support for Irregular and Loosely Synchronous Problems. Proceedings of the Conference on High Performance Computing for Flight Vehicles, 1992.

....parallelization of most applications. There are a number of partitioning algorithms available in the literature [1, 10, 14, 18, 19, 28] This list is by no means complete. In a large number of such problems the computational structure (or dependencies) can be constructed only during execution [5]. For such cases these graphs must be constructed at runtime; thus it is important that the partitioning of data be done at runtime. Achieving this in parallel is clearly necessary, else partitioning itself would become a bottleneck. This paper is focused on a subclass of applications in which the ....

....is limited to vertices that are physically proximate. Examples of such applications include finite element calculations [16] molecular dynamics [4] particle dynamics [23] particle in a cell [13, 27] region growing [7] and statistical physics [6] A list of other such applications is given in [5]. For these applications, partitioning can be achieved by exploiting the above property. Essentially, proximate points are clustered together and form a partition such that the number of points attached to each partition are approximately equal. Most of the interactions are local and the amount of ....

[Article contains additional citation context not shown here]

Alok Choudhary, Geoffrey C. Fox, Seema Hiranandani, Ken Kennedy, Charles Koelbel, Sanjay Ranka, and Joel Saltz. Software Support for Irregular and Loosely Synchronous Problems. In Proceedings of the Conference on High Performance Computing for Flight Vehicles, 1992. To appear.

....computational phases where patterns of data access and computational cost cannot be anticipated until runtime. In this class of problems, once runtime information is available, data access patterns are known before each computational phase. These problems are called irregular concurrent problems [9]. Examples of irregular concurrent problems include adaptive and self adaptive explicit, multigrid unstructured computational fluid dynamic solvers [29, 15] molecular dynamics codes (CHARMM [5] AMBER [43] GROMOS [40] etc. diagonal or polynomial preconditioned iterative linear solvers [41] ....

....due to reduction operations. The reduction operations in a Forall construct are specified using the Fortran D REDUCE construct. Reduction inside a Forall construct is important for representing a considerable set of scientific computations such as those found in sparse and unstructured problems [9]. This representation also preserves explicit parallelism available in the underlying computations. 3 Communication Schedule Reuse The cost of carrying out an inspector (phases B, C and D in Figure 8) can be amortized when the information produced by the inspector is computed once and then used ....

[Article contains additional citation context not shown here]

A. Choudhary, G. Fox, S. Hiranandani, K. Kennedy, C. Koelbel, S. Ranka, and J. Saltz. Software support for irregular and loosely synchronous problems. Computing Systems in Engineering, 3(1-4):43--52, 1992. Papers presented at the Symposium on High-Performance Computing for Flight Vehicles, December 1992.

No context found.

Choudhary, A., Fox, G., Ranka, S., Hiranandani, S., Kennedy, K., Koelbel, C., and Saltz, J. "Software support for irregular and loosely synchronous problems," Computing Systems in Engineering, 3(1--4):43--52, 1992. CSE-MS 118, CRPC-TR92258.

.... Stage When Ties are Broken by Choosing Neighbor With Smallest ID Resolving Ties at Random: The region growing problem is a representative of a type of loosely synchronous problems, known as adaptive irregular problems, whose data objects evolve during the computation in a time synchronized manner [5]. The problem exhibits a dynamic behavior that starts with a high degree of parallelism that very rapidly diminishes to a much lower degree of parallelism. In order to increase the degree of parallelism in the algorithm, we introduced an element of randomness to our parallel implementations. ....

G. Fox et al, "Software support for irregular and loosely synchronous problems", Tech. Report, Northeast Parallel Architectures Center, Syracuse Univ., May 1992.

.... performed at the time of compilation by giving directives in the language to decompose the data and its corresponding computations (based on the owner computes rule) 3] For a large class of scientific problems that are irregular in nature, achieving a good mapping is considerably more difficult [4]. The nature of these irregularities may not be known at the time of compilation and can be derived only at runtime. The handling of such irregular problems requires runtime information to partition the computation in such a fashion that each processor recieves an approximately equal amount of ....

Alok Choudhary, Geoffrey C. Fox, Seema Hiranandani, Ken Kennedy, Charles Koelbel, Sanjay Ranka, and Joel Saltz. Software support for irregular and loosely synchronous problems. In Proceedings of the Conference on High Performance Computing for Flight Vehicles, 1992. to appear.

First 50 documents

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