Sundeep Prakas . Performance Prediction of Parallel Programs. P D t esis, University of Calif ornia Los Angels, 1996. ttp://pcl.cs.ucla.edu/projects/sesame/.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....dataf rom a physical system and a MicroGrid simulation model system heterogeneity, irregular network structure, a complex Grid sof tware environment, and the tra#c modeling required by the network environments. There a wide variety of network simulation tools, including thosef or parallel systems [11, 15, 10] and thosef or Internet style networks [2, 3, 9] The parallel system tools typically do not model the detailed structure in complex heterogeneous network environments. Thosef or Internet systems typically do not support live tra#c simulation. Also, since Internet workloads are typically large ....

Sundeep Prakas . Performance Prediction of Parallel Programs. P D t esis, University of Calif ornia Los Angels, 1996. ttp://pcl.cs.ucla.edu/projects/sesame/.

....kernel: Execution of a discrete event model requires a kernel that can execute the events in their timestamp order. The simulation kernel provides sequential simulation for use on a uniprocessor, as well as parallel simulation using multiple synchronization protocols for a parallel architecture [Pra96]. ffl MPISIM: The simulation of MPI IO commands requires the ability to simulate MPI communication commands. MPISIM is a multi threaded MPI simulator that was developed for this purpose [Pra96] ffl PIO SIM: Simulates a subset of the individual and collective I O constructs provided by MPI IO. ....

....as well as parallel simulation using multiple synchronization protocols for a parallel architecture [Pra96] ffl MPISIM: The simulation of MPI IO commands requires the ability to simulate MPI communication commands. MPISIM is a multi threaded MPI simulator that was developed for this purpose [Pra96]. ffl PIO SIM: Simulates a subset of the individual and collective I O constructs provided by MPI IO. PIO SIM supports all operations except calls using the shared file pointer. This subset includes positioning using explicit 24 offsets and individual file pointers, independent and collective ....

[Article contains additional citation context not shown here]

Sundeep Prakash. Performance Prediction of Parallel Programs. Ph.d. dissertation, Computer Science Dept, UCLA, Los Angeles, CA, November 1996.

....author, no reported work on MPI simulation extends up to 512 or even to 128 processors. Howell s simulation results achieved within a factor of two for communication dominated program and a factor of ten for computation dominated [35] The maximum processor size used is 32. The MPISIM by Prakash [63] achieved within 20 of the target execution time. However, the processor range is up to 16 only. The two mentioned works (Howell s and Prakash s) are both full parallel simulation engine for MPI. The first is for T3D while the latter is for IBM SP2. 4.4.3 Validation of GDEM with Published ....

S. Prakash. Performance Prediction of Parallel Programs. PhD thesis, University of California - Los Angeles, November 1996.

....complex systems are analytically intractable. Although simulation is a widely applicable tool, its major limitation is its extremely long execution time for large scale systems. A number of simulators, including Parallel Proteus [22] LAPSE [13] SimOS [30] Wisconsin Wind Tunnel [28] and MPI SIM [26] have been developed to control the execution time of simulation models of parallel programs. The simulators typically use direct execution to reduce the cost of simulating sequential instructions and use parallel discrete event simulation to exploit parallelism within the simulator to reduce the ....

....a set of core non blocking MPI functions [27] The interconnection network model currently ignores contention in the network. More detailed models are being developed, but given the excellent validation obtained with the simpler model for a variety of benchmarks both here and in previous work [26], this was not considered to be a serious limitation. The parallel I O component of COMPASS simulates the individual and collective I O constructs provided by MPI IO. These constructs include creating, opening, closing and deleting a file; most data access (read write) operations; and a local ....

[Article contains additional citation context not shown here]

S. Prakash. "Performance Prediction of Parallel Programs," Computer Science Dept, UCLA, Ph.D. thesis, 1996.

....as well as parallel simulation using multiple synchronization protocols for a parallel architecture. ffl MPISIM: The simulation of MPI IO commands requires the ability to simulate MPI communication commands. MPISIM is a multi threaded MPI simulator that was developed for this purpose [Pra96] ffl PIO SIM: simulates the individual and collective I O constructs provided by MPI IO. These constructs include creating, opening, closing and deleting a file; most data access (read write) operations; and a local datatype constructor introduced as part of the MPI IO specification. The ....

....simulated by a set of disk models. We have included simple models based on seek time, rotational latency, and data transfer rate as well as a higly detailed model developed at Dartmouth [NK96] Since we restrict our attention in to parallel I O simulations, the interested reader is referred to [Pra96] for information on MPISIM. The I O simulator has been designed to be both modular and extensible: it is relatively easy to replace individual modules at each of the preceding levels. In particular, it is straightforward to replace the SSD models, to modify the caching or partitioning policies ....

Sundeep Prakash. Performance Prediction of Parallel Programs. Ph.d. dissertation, Computer Science Dept, UCLA, Los Angeles, CA, November 1996.

....is translated to a call to the corresponding MPI SIM function. MPI SIM internally implements each call to a collective function in terms of the core communication commands described in Section 2.1. For brevity, we do not describe the translation in the paper; the reader is referred to Prakash (1996). We briefly describe the simulation of the core commands. The sends in the MPI core are simulated by sending a message (with source, destination, tag, communicator and data) to the receiver LP. The message is timestamped with the send timestamp, which is the current simulation time of the sending ....

....Message Protocol (NMP) Chandy and J. Misra 1979) the Conditional Event Protocol (CEP) Chandy and Sherman 1989) and a new protocol, which is a combination the two (Jha and Bagrodia 1993) Due to space limitations, we have omitted details of the protocol; the interested reader is referred to Prakash (1996). The primary overhead in implementing parallel conservative protocols is due to the communications to compute EIT and the blocking suffered by an LP that has not been able to advance its EIT. We have suggested and implemented a number of optimizations to significantly reduce the frequency and ....

Prakash, S. Performance Prediction of Parallel Programs. Ph.D. Thesis, Computer Science Department, Ucla, Los Angeles, CA 90095, November 1996.

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