Bagrodia R., Prakash S.: MPI-SIM: Using Parallel Simulation to Evaluate MPI Programs. In Proc. of Winter Simulation Conference, 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....[10] optimizes MPI s collective communication primitives for clusters connected through a wide area network. In MPI FM [11] and MPI AM [19, 12] they attempt to optimize the performance of lower level communication devices for MPI. Their techniques can be applied to our TMPI system. MPI Lite [15], LPVM [20] and TPVM [6] study the problem of running message passing programs using threads on a single sharedmemory machine. To our knowledge, there is no research e#ort towards running MPI nodes using threads on SMP clusters. Our research complements the above work by focusing on taking ....

S. Prakash and R. Bagrodia. MPI-SIM: using parallel simulation to evaluate MPI programs. In Proceedings of Winter simulation, pages 467--474, Washington, DC., December 1998.

....without changing the programming interface. Most of previous MPI researches are ACM Transactions on Programming Languages and Systems, Vol. 0, No. 0, January 2000. 1002 Hong Tang et al. focused on distributed memory machines or workstation clusters, e.g. Bruck et al. 1997] The MPI SIM project [Prakash and Bagrodia 1998] has used multi threading to simulate MPI execution on distributed memory machines for performance prediction as we will discuss in Section 3.1. Thread safety of MPI systems is addressed in [MPI Forum 1999; Protopopov and Skjellum 1998; Skjellum et al. 1996] and recent commercial MPI products from ....

....a function pointer and it may call foo( with argument x or call foo2( without any extra argument. As a result, it is very hard, if not impossible, for pointer analysis to predict whether foo3( should carry an additional argument in executing f( The second solution, which is used by MPI SIM [Prakash and Bagrodia 1998], is called array replication. The preprocessor re declares each permanent variable with an additional dimension, whose size is equal to the total number of MPI nodes. There are three problems with this approach. First, the number of threads cannot be determined in advance at compile time. MPI SIM ....

[Article contains additional citation context not shown here]

PRAKASH, S. AND BAGRODIA, R. 1998. MPI-SIM: using parallel simulation to evaluate MPI programs. In Proceedings of Winter simulation. Washington, DC., 467--474.

....[38] and TPVM [15] projects. Both projects do not address how a PVM program can be executed in a multi threaded environment without changing the programming interface. Most of previous MPI research is focused on distributed memory machines or workstation clusters, e.g. 10] The MPI SIM project [8] has used multi threading to simulate MPI execution on distributed memory machines for performance prediction as we will discuss in Section 3.1. Thread safety is addressed in [3, 30, 33] However, their concern is how multiple threads can be invoked in each MPI node, but not how to execute each ....

....function pointer and it may call foo( with argument x or call foo2( without any extra argument. As a result, it is very hard, if not impossible, for pointer analysis to predict whether foo3( should carry an additional argument in executing f( The second solution, which is used by MPI SIM [8], is called array replication. The preprocessor re declares each permanent variable with an additional dimension, whose size is equal to the total number of threads. There are several problems with this approach. First, the number of threads cannot be determined in advance at compile time. MPI SIM ....

[Article contains additional citation context not shown here]

R. Bagrodia and S. Prakash. MPI-SIM: Using Parallel Simulation to Evaluate MPI Programs. In Proc. of Winter Simulation Conference, 1998.

....with overhead higher than our single layer scheme. In terms of lock free management, our scheme is more sophisticated for larger concurrency and better efficiency since queues in TMPI can be of arbitrary lengths and allow concurrent accesses between a sender and a receiver. The MPI Lite project [6] has used multi threaded execution to simulate MPI execution and their system is intended for performance prediction when the desired configuration is not available. Thus its communication library is based on the discrete event simulation theory. MPI Lite eliminates the global variables of an MPI ....

....whether foo3( should carry an additional argument in executing f( ffl Array replication. Another solution is to use array replication that redeclares each permanent variable with an additional dimension, whose size is equal to the total number of threads. This solution is used by MPI Lite [6]. There are a number of problems with this approach. First, the number of threads cannot be determined in advance at compile time. MPI Lite uses an upperlimit to allocate space and thus the space cost may be excessive. Second, even though the space of global variables could be allocated ....

R. Bagrodia and S. Prakash. MPI-SIM: Using Parallel Simulation to Evaluate MPI Programs. In Proceedings of 1998 Winter Simulation Conference, 1998.

....in a multi threaded environment without changing the programming interface. Most of gave up this design due to insufficient address space and software incompatibility [27] previous MPI research is focused on distributed memory machines or workstation clusters, e.g. 9] The MPISIM project [6, 7] has used multi threading to simulate MPI execution on distributed memory machines as we will discuss in Section 3.1. Thread safety is addressed in [3, 26, 29] However, their concern is how multiple threads can be invoked in each MPI node, but not how to execute each MPI node as a thread. These ....

....variables are passed to functions that need to access them. There is no overhead other than parameter passing, which can usually be done quite efficiently. The problem is that such an approach is not general and the transformation could fail for some cases. The second solution, which is used in [7], is called array replication. The preprocessor re declares each permanent variable with an additional dimension, whose size is equal to the total number of threads. There are several problems with this approach. First, the number of threads cannot be determined in advance at compile time. ....

[Article contains additional citation context not shown here]

R. Bagrodia and S. Prakash. MPI-SIM: Using Parallel Simulation to Evaluate MPI Programs. In Proc. of Winter Simulation Conference, 1998.

....[34] and TPVM [12] projects. Both projects do not address how a PVM program can be executed in a multi threaded environment without changing the programming interface. Most of previous MPI research is focused on distributed memory machines or workstation clusters, e.g. 9] The MPI SIM project [6, 7] has used multithreading to simulate MPI execution on distributed memory machines as we will discuss in Section 3.1. Thread safety is addressed in [3, 26, 29] However, their concern is how multiple threads can be invoked in each MPI node, but not how to execute each MPI node as a thread. These ....

....pointer and it may call foo( with argument x or call foo2( without any extra argument. As a result, it is very hard, if not impossible, for pointer analysis to predict whether foo3( should carry an additional argument in executing f( The second solution, which is used by MPI Lite [7], is called array replication. The preprocessor re declares each permanent variable with an additional dimension, whose size is equal to the total number of threads. There are several problems with this approach. First, the number of threads cannot be determined in advance at compile time. MPI SIM ....

[Article contains additional citation context not shown here]

R. Bagrodia and S. Prakash. MPI-SIM: Using Parallel Simulation to Evaluate MPI Programs. In Proc. of Winter Simulation Conference, 1998.

....code by using symbolic expressions parameterized by direct measurement. More sophisticated performance estimation for these sequential fragments is possible, but we do not do so here. In order to demonstrate the impact of these optimizations, we have combined the MPISim parallel simulator [6, 26, 27, 28] with the dHPF compiler infrastructure [2] into a program simulation framework that incorporates the new techniques described above. The MPI Sim simulator simulates unmodified MPI programs and uses both direct execution and parallel simulation to achieve substantial reductions in simulation time. ....

....2. RELATED WORK Because analytical performance prediction can be intractable for complex applications, program simulations are commonly used for such studies. It is well known that simulations of large systems tend to be slow. To improve the simulators, direct execution has been used [21, 27, 29]. Direct execution simulators make use of available system resources to directly execute portions of the application code and simulate architectural features that are of specific interest, or are unavailable. For example, simulators can be used to study various architectural components such as the ....

[Article contains additional citation context not shown here]

S. Prakash and R. L. Bagrodia. Mpi-sim: using parallel simulation to evaluate mpi programs. In Proceedings of IEEE Winter Simulation Conference, Washington, DC, 1998.

....code by using symbolic expressions parameterized by direct measurement. More sophisticated performance estimation for these sequential fragments is possible, but we do not do so here. In order to demonstrate the impact of these optimizations, we have combined the MPISim parallel simulator [6, 26, 27, 28] with the dHPF compiler infrastructure [2] into a program simulation framework that incorporates the new techniques described above. The MPI Sim simulator simulates unmodified MPI programs and uses both direct execution and parallel simulation to achieve substantial reductions in simulation time. ....

....2. RELATED WORK Because analytical performance prediction can be intractable for complex applications, program simulations are commonly used for such studies. It is well known that simulations of large systems tend to be slow. To improve the simulators, direct execution has been used [21, 27, 29]. Direct execution simulators make use of available system resources to directly execute portions of the application code and simulate architectural features that are of specific interest, or are unavailable. For example, simulators can be used to study various architectural components such as the ....

[Article contains additional citation context not shown here]

S. Prakash and R. L. Bagrodia. Mpi-sim: using parallel simulation to evaluate mpi programs. In Proceedings of IEEE Winter Simulation Conference, Washington, DC, 1998.

....has been validated against several MPI implementations including those on the IBM SP and SGI Origin 2000 [6] The simulation kernel provides support for sequential and parallel execution of the simulator. Parallel execution is supported via a set of conservative parallel simulation protocols [20], which typically work as follows: Each application process in the simulation is modeled by a Logical Process (LP) Each LP can execute independently, without synchronizing with other LPs, until it executes a wait operation (such as an MPI Recv, MPI Barrier, etc) a synchronization protocol is ....

S. Prakash and R. L. Bagrodia, "MPI-SIM: using parallel simulation to evaluate MPI programs," IEEE WSC, 1998.

....conclusions. 2 Related Work Because analytical performance prediction can be intractable for complex applications, program simulations are commonly used for such studies. It is well known that simulations of large systems tend to be slow. To improve the simulators, direct execution has been used [20, 26, 28]. Direct execution simulators make use of available system resources to directly execute portions of the application code and simulate architectural features that are of specific interest, or are unavailable. For example, simulators can be used to study various architectural components such as the ....

....multiple parallel architectures including a distributed memory IBM SP2 as well as a sharedmemory SGI Origin 2000. The simulation kernel provides support for sequential and parallel execution of the simulator. Parallel execution is supported via a set of conservative parallel simulation protocols [26], which typically work as follows: Each 2 In the future, we plan to synthesize this information for existing MPI codes as well. The dHPF infrastructure supports very general computation partitioning, communication analysis, and symbolic analysis capabilities that make this feasible for a wide ....

[Article contains additional citation context not shown here]

S. Prakash and R. L. Bagrodia, "MPI-SIM: using parallel simulation to evaluate MPI programs," Proceedings of IEEE Winter Simulation Conference, Washington, DC, USA, 1998.

No context found.

Prakash, S. and R.L. Bagrodia. MPI-SIM: using parallel simulation to evaluate MPI programs. in Proceedings (Cat. No.98CH36274) Proceedings of IEEE Winter Simulation Conference. 1998. Washington, DC, USA: IEEE.

No context found.

Bagrodia R., Prakash S.: MPI-SIM: Using Parallel Simulation to Evaluate MPI Programs. In Proc. of Winter Simulation Conference, 1998.

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