| Jack B. Dennis, Guang-Rong Gao, and Kenneth W. Todd. Modeling the weather with a data flow supercomputer. IEEE Transactions on Computers, 33(7):592--603, July 1984. |
....Data flow programs also exhibit fine granularity of parallelism and fine communication granularity: Data flow tokens are equivalent to short messages. Data flow actors (and macro actors) correspond to short code blocks. The appeal of data flow systems is their ability to support software pipelining[4]. Such pipelining results in the frequent exchange of short messages. Certain numerical matrix algorithms have been developed that are not easily vectorizable. Such algorithms rely on the frequent exchange of single word messages during the course of the computation [5] Similar behaviour is ....
J. B. Dennis, G.-R. Gao, and K. W. Todd, "Modeling the Weather with a Data Flow Supercomputer, " IEEE Trans. Comput., Vol. C-33, No. 7, July 1984. pp. 592-603,
....at five times the processor clock rate, so there is little communication latency. Sigma 1 and EM 4 are available in large configurations, but the only programs that have been run are small, regular, and hand tuned. Static dataflow machines have been applied primarily to regular, structured problems[10, 11, 24]. Dataflow languages, including Sisal and a restricted form of Id, have been implemented on conventional architectures, by exploiting the regular structure in traditional FORTRAN like applications[S, 19] Our work attempts to fill some of this empirical vacuum by implementing Id90 on large ....
J. B. Dennis, G. Gao, and K. W. Todd. Modeling the weather with a data flow supercomputer. IEEE Transactions on Computers, c-33(7):592 603, July 1984.
....Also dynamic and irregular applications might cause excessive waste of cycles when mapped to a blocking thread model. 142 8. 2 Software Multithreaded Systems In the classical strict data flow model of computation, an instruction is enabled for execution when all its operands are available [66, 85, 63, 65, 68, 47, 155, 70, 77, 127, 130, 86, 123, 125, 12, 97, 133, 132, 124, 15, 17, 45, 57, 150, 140]. To enforce the enabling condition, the instructions that produce such operands must be able to send a synchronization signal to all the instructions that will consume the recently produced result. This model proved unyielding for the implementation of machines based on current standard ....
Jack B. Dennis, Guang-Rong Gao, and Kenneth W. Todd. Modeling the weather with a data flow supercomputer. IEEE Trans. on Computers, 33(7):592--603, Jul. 1984.
....want to distribute the matrix elements in a data parallel fashion. However, the function boundaries in the MTA code do not have to correspond to function boundaries in the source code. For instance, one of the easiest data parallel applications to program on a parallel machine is a mesh simulation [8], in which a physical region is represented as a grid of points, and at each time step, each grid point is updated by performing a function on its previous state and the state of its nearest neighbors. Each time step could be done in the source code with a single function call or series of ....
Jack B. Dennis, Guang-Rong Gao, and Kenneth W. Todd. Modeling the weather with a data flow supercomputer. IEEE Transactions on Computers, 33(7):592--603, July 1984.
....invocations to execute in parallel, it is most suited for applications that require speed and or are intensively number crunching, such as mathematical software and real time simulations. An example of global weather model was implemented on Dennis s architecture using the functional language VAL [2]. A 20 fold improvement in performance was reported for that weather simulation application. Since Dennis s model suffers some of the drawbacks discussed in this paper, and that our system attempts to overcome, we feel confident that our system can achieve, at least similar and probably ....
Dennis, J.B., Gao, G.R., and Todd, K.W. Modeling the Weather with a Data Flow Supercomputer. IEEE Transactions on Computers, C-33, July 1984, 592-603.
....[8] have compared the implementation of the hydrodynamics modeling application SIMPLE in Id and in Fortran; Arvind, et.al [7] have also explored the benefits of finegrained parallelism in scientific applications. In one of the original papers about programming for dataflow machines, Dennis, et.al. [16] describe the process of implementing a weather modeling code for a theoretical static dataflow architecture as well as the sources of parallelism within the code. Other related work includes high performance implementations of functional languages. Most notably, SISAL has shown very good ....
DENNIS, J. B., GAO, G. R., AND TODD, K. W. Modeling the weather with a data flow supercomputer. IEEE Transactions on Computers C-33, 7 (July 1984), 592--603.
No context found.
Jack B. Dennis, Guang-Rong Gao, and Kenneth W. Todd. Modeling the weather with a data flow supercomputer. IEEE Transactions on Computers, 33(7):592--603, July 1984.
....slots which count how many 5 DATA DATA SYNC DATA SYNC BROADCAST REDUCE REDUCE Figure 6. Building Block for ReductionBroadcast Tree input values have arrived and trigger the execution of a fiber when all required inputs are ready. It is in the style of a classical static dataflow actor [6]. The core of the building block is a pair of fibers, one for reduction and one for broadcast. When a child produces a value, it uses the EARTH DATA SYNC operator to pass that value to one of the input locations and synchronize the sync slot. Note that both children in Figure 6 use the same ....
Jack B. Dennis, Guang-Rong Gao, and Kenneth W. Todd. Modeling the weather with a data flow supercomputer. IEEE Transactions on Computers, 33(7):592--603, July 1984.
....dataflow computers have no program counter or other form of centralized sequential control mechanism. Consequently, the order of instruction execution is restricted only by data dependencies within the dataflow programs. In recent years much research has been conducted on dataflow architectures [1, 11, 14, 13, 32]. Two major approaches to the architectures of dataflow computers have been proposed: the static dataflow model [15] and the tagged token model [5, 35] The dynamic model has evolved into the development of the modern dynamic dataflow machines such as the Monsoon dataflow machine at MIT and ....
J. B. Dennis, G. R. Gao, and K. W. Todd. Modeling the weather with a data flow supercomputer. IEEE Transactions on Computers, C-33(7):592--603, 1984.
....dynamic dataflow principles and the requirements of a general purpose functional programming language, and is described in Section 6. Early proposals for dataflow computers suggested that machine instructions would be in (nearly) one to one correspondence with the actors of dataflow program graphs [48, 30, 42, 47]. The design discussed here departs from this view. Instead of designing hardware to mimic data flow graphs, the objective is to define hardware structures that can efficiently implement the semantic content of dataflow program graphs [41] The architecture departs from a direct implementation of ....
Jack B. Dennis, Guang-Rong Gao, and Kenneth W. Todd, "Modeling the weather with a data flow supercomputer," IEEE Transactions on Computers, 33(7):592--603, July 1984.
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