T. Decker. Virtual Data Space - A universal load balancing scheme. In G. Bilardi, A. Ferreira, R. Lling, and J. Rolim, editors, Proc. of the 4th International Symp. on Parallel Algorithms for Irregularly Structured Problems, volume 1253 of Lecture Notes in Computer Science, pages 159166. Springer, 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....[1,5] 1,10] 1,20] 1,5] 1,10] 1,20] 4 3 3 3 2 2 2 8 11 12 14 5 6 6 16 16 18 20 7 6 7 32 31 35 37 10 13 14 VI. FINAL REMARKS A large number of tools and environments that support the development of parallel applications have been proposed and implemented [CJ99] [Dec97], FMD98] Wei99] CJ99] presents the PMESC programming tools. PMESC is a library that offers routines typically used in SPMD applications. SAMBA allows the resuse not only of specific routines, but of the whole SPMD structure. Also, one of SAMBA s main goals is to support the choice of an ....

....offers routines typically used in SPMD applications. SAMBA allows the resuse not only of specific routines, but of the whole SPMD structure. Also, one of SAMBA s main goals is to support the choice of an appropriate load balancing strategy. PMESC does not deal with this issue. The VDS library [Dec97] offers mechanisms for the distribution of tasks, with the goal of balancing the load among the participating processors in a parallel application. However, the library does not offer alternative strategies. Once the type of application to be developed is defined, VDS adopts one out of two ....

T. Decker, "Virtual Data Space - A universal load balancing scheme", LNCS 1253 (1997), 159--166.

....Designing a scheduler to achieve all of the above goals is not a trivial task. There are three main performance parameters for scheduling algorithms for multithreaded computations, namely their space complexity, their execution time and the communication cost incurred by them. Several systems [16, 23, 27, 30, 35] have used work stealing to achieve the scheduling goals described above and to provide high performance. Work stealing is a technique in which underutilized processors try to steal work from other, hopefully overutilized processors. Indeed, work stealing has been proved (see e.g. 1, 8, 12, ....

....The algorithm presented in [12] has been implemented in CILK [8, 27] which is a C based language for programming multithreaded computations. Several lightweight and ecient userlevel POSIX threads (or Pthreads [32] packages are available for shared memory programming 44 (see e.g. 43, 53] VDS [23] is a load balancing tool which employs randomized work stealing techniques as its quantitative load balancing mechanism. An overview of existing tools that incorporate load balancing techniques and support di erent programming paradigms is provided in [26] 6 Discussion and Directions for ....

T. Decker, \Virtual Data Space - A Universal Load Balancing Scheme," Proceedings of the 4th International Symposium on Solving Irregularly Structured Problems in Parallel, pp. 159-166, Paderborn, Germany, 1997. 47

No context found.

T. Decker. Virtual Data Space - A universal load balancing scheme. In G. Bilardi, A. Ferreira, R. Lling, and J. Rolim, editors, Proc. of the 4th International Symp. on Parallel Algorithms for Irregularly Structured Problems, volume 1253 of Lecture Notes in Computer Science, pages 159166. Springer, 1997.

....as slow as a single run with the optimal precision. This is due to the fact that the algorithm realizes quite early that the chosen precision is too low; it then quits, doubles the precision, and starts again. 5. 2 Parallel aspects Our implementation is based on the load balancing system VDS [Dec97] which allowed the integration of the various dag scheduling schemes and the central scheduling algorithm. Furthermore, the system made it possible to generate the execution profiles. Figure 12 shows two execution profiles (excerpts) of the parallel Descartes method for a Chebyshev polynomial of ....

T. Decker. Virtual Data Space---A universal load balancing scheme. In G. Bilardi, A. Ferreira, R. Luling, and J. Rolim, editors, Proceedings of the 4-th International Symposium on Solving Irregularly Structured Problems in Parallel, volume 1253 of Lecture Notes in Computer Science, pages 159--166. Springer, 1997.

....for each hierarchy level. VDS introduces the concept of dynamic processor groups that allows the assignment of processors to parallel sub tasks. The assignment is changed automatically as soon as the computation in one of the groups terminates. An earlier version of VDS has been presented in [18]. The present version 2.0 is available for download [17] The load balancing techniques and the control mechanisms that are needed in order to obtain an e cient parallelization depend on the application type. For example, if the tasks of the applications incorporate precedence constraints that ....

T. Decker. Virtual Data Space - A universal load balancing scheme. In G. Bilardi, A. Ferreira, R. Lling, and J. Rolim, editors, Proc. of the 4th International Symp. on Parallel Algorithms for Irregularly Structured Problems, volume 1253 of Lecture Notes in Computer Science, pages 159166. Springer, 1997. 39

....with load balancing problems are described in the following section. IMPLEMENTATION PLATFORM The implementation of the parallel algorithm is based upon Virtual Data Space (VDS) developed at the University of Paderborn. VDS is a general load balancing library for irregular applications [1, 2]. The library shields the programmer from the need to implement functions that do not strictly concern her application. Load balancing, termination detection, and data management are done completely automatically. VDS supports various parallel programming paradigms such as parallel independent ....

Decker, T. (1997). Virtual Data Space - A universal load balancing scheme. Proc. 4th IRREGULAR, 159--166.

....Search trees for Chebyshev polynomials are wide and deep. Mignotte polynomials are defined as An (x) x n 2(5x 1) 2 . Search trees for Mignotte polynomials are narrow and extremely deep. Figure 4 shows examples of search trees. Our implementation is based on the load balancing system VDS [Dec97] We integrated the various scheduling methods for DAGs and the phase parallel scheduler. We also used the system to generate execution profiles. Fig. 5. Execution profiles of a Taylor shift (left) and of a root isolation (right) The horizontal axis is the time axis (the time is given in ....

T. Decker. Virtual Data Space---A universal load balancing scheme. In G. Bilardi, A. Ferreira, R. Luling, and J. Rolim, editors, Proceedings of the 4-th International Symposium on Solving Irregularly Structured Problems in Parallel, volume 1253 of Lecture Notes in Computer Science, pages 159--166. Springer, 1997.

....real roots noting that real root isolation can be used as a subalgorithm in the isolation of nonreal complex roots. Our implementation uses the Daisy library (Data Management Library for Parallel Systems) 6] which consists of two cooperating parts, the load balancer VDS (Virtual Data Space) [5], and the memory manager Diva (Distributed Variables) for the provision of global variables. In contrast to load balancing systems based on process migration, VDS describes tasks by data items This work was supported by the DFG Sonderforschungsbereich 376 : Massive Parallelit t Algorithmen, ....

T. Decker. Virtual Data Space - A universal load balancing scheme. In Proc. of the 4th International Symp. on Parallel Algorithms for Irregularly Structured Problems, pages 159--166, 1997.

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