No full-text available
To read the full-text of this research,
you can request a copy directly from the author.
Abstract
First Page of the Article
... begin For all input channels while a message is pending and next-in = t-safe Read message on channel next-in = I If the message is not a null message Insert message into event list end if end while end for end [3][4][5][6][7][8][9][10][11][12] To ensure that deadlock is successfully avoided as proposed by the Chandy-Misra paradigm, Mannix outlines two null message conditions that must hold in the simulation: ...
... Describing a message channel (ij) for communicating logical processes LP and LPj,[3][4][5][6][7][8][9][10][11][12][13][14] for each message transmitted over channel (ij), LPi calculates the lower bound for the next output message as outlined above and appends this time estimate to the message.All messages sent between LP and LPj are of the form [t, m], where t is the lower bound estimate and m is the message, null or event, transmitted by LPi. Note that t is distinct from the timestamp tmsg of the event message, which is included in m. ...
The primary goal of distributed discrete event simulations is to achieve speedup in simulation execution time by distributing the processing of the simulation over multiple processors. When partitioned for distribution in this fashion, simulations are typically partitioned such that there are more processes than processors. This thesis reviews existing methods for distributed discrete event simulations, and proposes general guidelines for efficient partitioning for a given communications topology based on empirical evidence. A performance analysis is conducted for two approaches to partitioning the system. The first method chosen is a mapping of multiple processes to a processor and the second approach utilizes a distributed event list approach, developed by Mannix. This approach combines smaller processes into a larger single process, incorporating a next event list similar to that used in a sequential simulation. Empirical studies compare the performance of the two approaches under a variety of conditions. The traditional Chandy-Mirsa approach to system partitioning is demonstrated to yield overall better performance than the distributed event list algorithm. General guidelines for partitioning the system for both approaches are developed based on the performance comparisons.
Virtual time is a new paradigm for organizing and synchronizing distributed systems which can be applied to such problems as distributed discrete event simulation and distributed database concurrency control. Virtual time provides a flexible abstraction of real time in much the same way that virtual memory provides an abstraction of real memory. It is implemented using the Time Warp mechanism, a synchronization protocol distinguished by its reliance on lookahead-rollback, and by its implementation of rollback via antimessages.
Amdahl's law predicts time reduction for a fixed problem size. If you instead apply P processors to a task that has serial fraction f, scaling the problem to take the same amount of time as before, the speedup is f + P(1–f) = P – f(P – 1) and the serial fraction f does not theoretically limit parallel speed enhancement if the workload scales in its parallel component.
Traditional discrete-event simulations employ an inherently sequential algorithm. In practice, simulations of large systems are limited by this sequentiality, because only a modest number of events can be simulated. Distributed discrete-event simulation (carried out on a network of processors with asynchronous message-communicating capabilities) is proposed as an alternative; it may provide better performance by partitioning the simulation among the component processors. The basic distributed simulation scheme, which uses time encoding, is described. Its major shortcoming is a possibility of deadlock. Several techniques for deadlock avoidance and deadlock detection are suggested. The focus of this work is on the theory of distributed discrete-event simulation.
This paper describes a software tool for parallel systems performance modeling, called Hypersim. Hypersim was developed by Intel Scientific Computers in order to evaluate the performance of various architectural alternatives for its future parallel machines. Hypersim runs on the Sun* 3 workstation and on the Intel iPSC concurrent super computer. The use of the iPSC system allows us to simulate large models, which do not fit within the Sun's physical memory and cause excessive paging delays.
Hypersim currently simulates hypercube architectures with various arities and dimensions. It implements the full user interface of the iPSC Node Operating System, which includes facilities for dynamic process creation and message passing between processes. The simulator is highly parametrized to represent timings for memory access, message and interrupt propagation, process switches, and instruction execution.
Key software and hardware functions are implemented in a modular manner, so that detail can be added as the architecture is elaborated. This object oriented approach results in a relatively simple and extensible design, and the iPSC version differs minimally from the Sun version. Global object management and synchronization are provided by the Interwork II concurrent programming environment.
We propose a distributed simulation method which is particularly well suited for the simulation of large synchronous networks. In general, distributed simulation has significant potential for alleviating the time and memory constraints often encountered when using conventional simulation techniques. Currently proposed methods for distributed simulation suffer performance degradation due to the employment of strategies for preventing deadlock and due to artificial blocking of processes. We describe a new method for doing distributed simulation which is deadlock free whenever the physical system being simulated is deadlock free. Furthermore, our method does not suffer from artificial blocking of processes.
rts to appoint their own expert witnesses instead of rel expertBj0j"2BLj"? nselL-0 do so. The probl of rel expertBj0j"2BLj"??2Blel disinterested reso rce acceptablesoB-""-BL-'2? The AAAS Demonstration Program seeks to provide an answer to that q estion. The AAAS is cal???2BlqB-'k?"BL-???2Blpro-Bj ch as SCSI, to participate in compilL-???2Blpro-Bj2z2'BL fficientlk ?kBL-j"?2Bl in compilL-???2Blpro-Bj2z2'BL-k' of the tr st req ired for s ch co rt appearances. The perts. It ales. TheBj0jkjBL0'?2Bl??2Blpro-Bj2z2'BL-k'? co rts seeking experts. It wil0'?2Bl??2Blpro-Bj2z2'BLing societies programs in s pport of this project. (For a broch re on the program, with the contact person's name and data, contact mcleod@sdsc.edu). Friday afternoon I attended the b siness meeting of Section T, Information, Comp ting, and Comm nica- tion. I noted that there were onlBlBL-j?2BlBlBL-k'?2B meeting that were sponsored by AAAS Section T, incl ding my Sat rday session. (Additional by AAA were "s
The Society for Computer Simulation
- W L Bain
- D S Scott
A Closer Look at Amdahl's Law
- C Moler