Greg Eisenhauer, Weiming Gu, Karsten Schwan, and Niru Mallavarupu. Falcon - toward interactive parallel programs: The on-line steering of a molecular dynamics application, August 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of material properties. In this paper, we describe our efforts in addressing the practical problems of working with very large molecular dynamics simulations. We have taken a computational steering approach in which simulation, data analysis, and visualization are combined into a single package [9,10,11,12]. However, we have adopted an approach that is extremely lightweight that is, it is memory efficient, simple to use, portable, easily extensible, and not dependent on expensive special purpose hardware. i.e. graphics workstations or high speed networking) The SPaSM Code Since 1992, we have ....

....large scale computing is still difficult, still takes hundreds of hours of computing time on the fastest machines available, and still generates an overwhelming amount of data. Some seem to believe that the data glut problem can be magically eliminated with very expensive special purpose hardware [10,11,12,16]. While this may be possible, our experience in the I WAY at Supercomputing 95 was a complete disaster (unless you consider showing a pre rendered MPEG movie of a MD simulation to a half dozen people a success) 17] Even if the real demo had worked, most scientists we know do not have the ....

[Article contains additional citation context not shown here]

G.Eisenhauer, W.Gu, K. Schwan, and N. Mallavarupu, "Falcon-Toward Interactive Parallel Programs : The On-line Steering of a Molecular Dynamics Application," Proc of the Third International Symposium on High Performance Distributed Computing (HPDC-3), IEEE Computer Society (1994), pg. 26-34.

....and steering. More comprehensive views on agents can be found in[3, 4] Most systems for monitoring distributed applications centralize the control over whether monitoring is activated or not, at a process representing the user. Examples of distributed monitoring systems are CUMULVS[6] Falcon[5], and Pablo[7] 6 Conclusions Future Work ACKNOWLEDGEMENTS This paper is based upon work supported in part by Boeing and the National Science Foundation Grant No. CDA 9619831. ....

G. Eisenhauer, W. Gu, K. Schwan, and N. Mallavarupu. Falcon - toward interactive parallel programs: The On-line steering of a molecular dynamics application. In Proceedings of High Performance Distributed Computing (HPDC-3), Aug. 1994.

....shown in Figure 2) it was executed, and produced the plot in Figure 3. Although our system continues to evolve, our preliminary feedback is very encouraging. Figure 3. Experimental Geometry of the Seismic Experiment 4. Related Work Several computational steering systems have been developed [3,4,6,9,10,15,16]. Some of them have focused on support for new applications which do not impose as many restrictions as legacy codes do. SCIRun [15] for example, uses an object oriented data flow approach to support runtime interaction for new applications in computational medicine. However, it does not easily ....

G. Eisenhauer, W. Gu, K. Schwan, and N. Mallavarupu, Falcon--Toward Interactive Parallel Programs: The On-line Steering of a Molecular Dynamics Application, Proc. Third Intl. Symposium on High-Performance Distributed Computing (HPDC-3), San Francisco, August 1994.

....includes the ability to select dynamically the parallel platforms to be used, to monitor the progress of the computational components, and to modify parameters as needed. Moreover, while this approach does not support the extensive system and performance monitoring capabilities of other approaches [9, 10, 11, 12, 13], it leverages existing tools and requires only modest effort on the part of the programmer. In short, it represents a potentially large improvement in functionality for only a small investment. This paper makes several contributions. First, it demonstrates the validity of this approach by ....

....facilities for monitoring and controlling computations during execution. These include systems primarily intended for controlling sequential programs on a single machine such as Dynascope [31] as well as systems more oriented towards remote control of scientific applications such as Falcon [9], VASE [11] and Paradyn [12] When compared to such systems, our approach focuses exclusively on making it easier to display intermediate and final results of a remote parallel computation rather than on collecting and displaying additional data such as system or application performance metrics. ....

G. Eisenhauer, W. Gu, K. Schwan, and N. Mallavarupu, `Falcon---towards interactive parallel programs: The on-line steering of a molecular dynamics application', Proceedings of the 3rd Symposium on High Performance Distributed Computing, San Francisco, April 1994, pp. 26--33.

.... changes are applied only when the actuator code is executed, at the point in the computation at which the actuator has been inserted into the code. However, such an approach provides no mechanism for the coordination of actions across multiple processes. Systems such as Pablo[10] Falcon[11], Progress[12] Magellan[5] and VASE[13] ensure consistency of steering actions through reliance on the placement of user or programmer defined instrumentation at safe points in the execution of the code. Progress provides a variety of instrumentation object types including sensors, actuators, ....

Greg Eisenhauer, Weiming Gu, Karsten Schwan, and Niru Mallavarupu, "Falcon - toward interactive parallel programs: The On-line steering of a molecular dynamics application," in Proceedings of High Performance Distributed Computing (HPDC-3), Aug. 1994.

....and for supporting the ability of scientists to interpret, control, and adjust the parameters of their applications in close to realtime. Recently, interactive steering techniques have been applied to scientific and engineering Fortran codes, image filtering[22] molecular dynamics simulation[4], and global atmospheric modeling code[28] To achieve these ends, researchers have developed toolkits and libraries such as Falcon[7] Progress[28] CUMULVS[6] and the VASE environment[11] Interactive systems such as these permit users to monitor and control, or steer, particular parameters of ....

G. Eisenhauer, W. Gu, K. Schwan, and N. Mallavarupu. Falcon - toward interactive parallel programs: The On-line steering of a molecular dynamics application. In Proceedings of High Performance Distributed Computing (HPDC3) , Aug. 1994.

....respectively. First, though, we briefly describe work related to DAQV II. 2 2 Related work DAQV is a confluence of several research ideas. Not surprisingly, certain aspects of its design, functionality, and implementation appear in other types of tools. For example, systems like Falcon [5], CUMULVS [7] VASE [14,26] and SCIRun [24] typically provide complete, closed environments for computational steering. At the core of these systems is functionality similar to that provided by DAQV II, but DAQV II is unique in that it implements computational steering middleware, which allows ....

G. Eisenhauer, W. Gu, K. Schwan, and N. Mallavarupu, Falcon -Toward Interactive Parallel Programs: The On-line Steering of a Molecular Dynamics Application, Proc. Third Intl. Symposium on High-Performance Distributed Computing (HPDC-3), San Francisco, August 1994.

....and guiding a long running program. The program may have a parallel implementation, but it need not [11] Typically, the role of user guidance is to improve the performance of the program, e.g. by redistributing data to improve locality [12] or by migrating work to improve load balancing [2]. Our work, in contrast, is not intended to expose aspects of an implementation that are normally hidden from the user, but rather to help the user create and manage many high level tasks for the implementation to execute. We begin in Section 2 by discussing how to design parallel interfaces that ....

G. Eisenhauer, W. Gu, K. Schwan, and N. Mallavvarupu. Falcon -- toward interactive parallel programs: The on-line steering of a molecular dynamics application. In Proceedings of the Third International Symposium on HighPerformance Distributed Computing, August 1994.

.... performance compared to SDDF [1] 2) events transported using the DataExchange communication infrastructure detailed in [3] 3) information flow, task execution rate, and task execution timings monitored using an on line monitoring facility (examples of such are W3 [7] Chaos MON, and Falcon [4]) and (4) an on line steering facility [4] 13] allowing ILI to affect the operation of its tasks. 5. Heuristics ILI s on line adaptation heuristics are applied cyclically in three different steps: 1) Detect State Phase which determines if all tasks are currently meeting their QoS constraints and ....

.... events transported using the DataExchange communication infrastructure detailed in [3] 3) information flow, task execution rate, and task execution timings monitored using an on line monitoring facility (examples of such are W3 [7] Chaos MON, and Falcon [4] and (4) an on line steering facility [4], 13] allowing ILI to affect the operation of its tasks. 5. Heuristics ILI s on line adaptation heuristics are applied cyclically in three different steps: 1) Detect State Phase which determines if all tasks are currently meeting their QoS constraints and if not, where the violations are, 2) ....

G. Eisenhauer, K. S. Weiming Gu, and N. Mallavarupu. Falcon - toward interactive parallel programs: the on-line steering of a molecular dynamics application. In Proceedings of High-Performance Distributed Computing 3. IEEE, August 1994.

.... The goal of adapting the introspection system is to improve the component s efficiency for a given execution instance: tuning the granularity of sensors, selecting the right prediction utility for the given monitoring stream s characteristics, reducing the monitoring overheads and perturbations [9, 24], choosing the right adaptation at a given point in time, and dealing with unpredictable disasters [17] Clearly, all this needs to be done without making the adaptive introspective component so heavyweight as to deny its benefits. 3 Implementation Details Currently, AIMS nodes report ....

Greg Eisenhauer, Weiming Gu, Karsten Schwan, and Niru Mallavarupu. Falcon - toward interactive parallel programs: The on-line steering of a molecular dynamics application, August 1994.

.... les across the computational grid, support scienti c collaboration by remote visualization of their data[18, 19] and interact with other computations via the Grid [1, 2] Unless run time monitoring is used to determine the appropriate and dynamic allocation of cluster resources to applications[6, 10], high performance is unlikely to be attained. Run time monitoring mechanisms are required to dynamically diagnose the performance of cluster programs. The monitoring tools commonly available to cluster programmers, however, are not only used for adjusting programs[22,11] at runtime, but they are ....

....frequencies and rates, thereby altering the precision vs. perturbation induced by monitoring. Flexible and dynamic analysis. It is well known that monitoring data should be condensed and ltered as close as possible to its points of capture, to reduce monitoring overheads and perturbation[20, 6]. However, the actual analyses to be performed typically depend on what monitoring is currently used for, and such analyses vary in their behavior, some causing little perturbation, others requiring substantial trace data before they may be applied. No single builtin set of analysis routines will ....

[Article contains additional citation context not shown here]

Greg Eisenhauer, Weiming Gu, Karsten Schwan and Niru Mallavarupu. "Falcon { Toward Interactive Parallel Programs: The On-line Steering of a Molecular Dynamics Application", In Proceedings of The Third International Symposium on High-Performance Distributed Computing (HPDC-3), San Francisco, August 1994. IEEE Computer Society

.... les across the computational grid, support scienti c collaboration by remote visualization of their data[21, 22] and interact with other computations via the Grid [1, 2] Unless run time monitoring is used to determine the appropriate and dynamic allocation of cluster resources to applications[7, 11], high performance is unlikely to be attained. Run time monitoring mechanisms are required to dynamically diagnose the performance of cluster programs. The monitoring tools commonly available to cluster programmers, however, are not only used for adjusting programs[25,12] at runtime, but they are ....

....frequencies and rates, thereby altering the precision vs. perturbation induced by monitoring. Flexible and dynamic analysis. It is well known that monitoring data should be condensed and ltered as close as possible to its points of capture, to reduce monitoring overheads and perturbation[23, 7]. However, the actual analyses to be performed typically depend on what monitoring is currently used for, and such analyses vary in their behavior, some causing little perturbation, others requiring substantial trace data before they may be applied. No single builtin set of analysis routines will ....

[Article contains additional citation context not shown here]

Greg Eisenhauer, Weiming Gu, Karsten Schwan and Niru Mallavarupu. "Falcon { Toward Interactive Parallel Programs: The On-line Steering of a Molecular Dynamics Application", In Proceedings of The Third International Symposium on High-Performance Distributed Computing (HPDC-3), San Francisco, August 1994. IEEE Computer Society

....scientists at unprecedented levels. Furthermore, in the late 80 s, researchers developed general approaches to program monitoring[1] 2] 3] followed by advances in program adaptation[4] and steering[5] Specifically, by coupling program monitoring and steering with online visualizations[6] [7] of scientific data, it has now become possible for scientists to interact with their simulations at runtime and in a variety of ways that do not depend on simulation designs or implementations. As a result, intermediate program values may be inspected at will, parameters may be changed on the ....

....In our future work, we will be adding steering infrastructure between the hydrology application and the CAVE5D and VizAD visualization tools. In other work, interactivity support has been integrated with high end visualization to provide steering of a parallel molecular dynamics application [7], Habanero [14] and Tango [15] provide collaborative support with Java based environments, and Cavernsoft provides collaboration support for Cave based visualizations[16] 3 Gigabit Ethernet Switch Gigabit Ethernet Switch multimedia workstations (SGI) Compute Servers leased lines personal ....

G. Eisenhauer, W. Gu, K. Schwan, and N. Mallavarupu, "Falcon - toward interactive parallel programs: The on-line steering of a molecular dynamics application," in Proc. Third Int'l Symp. on High-Performance Distributed Computing (HPDC-3). 1994, pp. 26--34, IEEE Computer Society.

....use of interactive high performance programs, on parallel and distributed execution platforms. More specifically, the Progress toolkit presented and evaluated in this paper provides facilities through which existing high performance multiprocessor programs are extended for increased interactivity [McCormick88, Jablanowski93, Eisenhauer94]. Once these extensions have been performed on a target application, a user can connect to her high performance program, extract application specific data regarding the program s execution state, perform steering actions, if desired, and then disconnect from the program. The runtime overheads ....

....analysis of results could both pinpoint errors in initial conditions and allow modifications to error convergence limits. CFD data is usually visualized for final analysis but with on line, interactive systems, various data might be visualized and modified throughout the simulation. MD [Eisenhauer94] is a molecular dynamics simulation that explores the statistical mechanics of complex liquids. The dominant computational requirement is the calculation of long range forces between particles. MD is an interesting steering target because load balancing is difficult and standard heuristics are not ....

Greg Eisenhauer, Weiming Gu, Karsten Schwan, and Niru Mallavarupu. "Falcon -- toward interactive parallel programs: The on-line steering of a molecular dynamics application." In Proceedings of The Third International Symposium on High-Performance Distributed Computing, San Francisco, CA, August 1994.

.... of performance motivated computational steering include the runtime adjustment of performance relevant parameters in real time applications [1] the adjustment of operating system mechanisms to improve application performance[2, 3] and the improvement of load balance in scientific applications [4]. Steering performed for purposes of experimentation with target applications alter models and applications attributes[5, 6, 4, 7] For example, the steering interface developed for the global atmospheric modeling code described in [8] permits space scientist to run the code with alternative ....

.... applications [1] the adjustment of operating system mechanisms to improve application performance[2, 3] and the improvement of load balance in scientific applications [4] Steering performed for purposes of experimentation with target applications alter models and applications attributes[5, 6, 4, 7]. For example, the steering interface developed for the global atmospheric modeling code described in [8] permits space scientist to run the code with alternative settings for atmospheric transport, to restart the code from a previously defined checkpoint, and to gradually inject certain chemical ....

[Article contains additional citation context not shown here]

G. Eisenhauer, W. Gu, K. Schwan, and N. Mallavarupu, "Falcon - toward interactive parallel programs: The on-line steering of a molecular dynamics application," in Proc. Third Int. Symp. on High-Performance Distributed Computing (HPDC-3), pp. 26--34, 1994.

....parallel applications and discusses ways in which a monitoring and steering system might benefit each. 2. 1 MD MD is an interactive molecular dynamics simulation developed at Georgia Tech in cooperation with a group of physicists exploring the statistical mechanics of complex liquids [XORL92, EGSM94] The specific molecular dynamics systems being simulated are n hexadecane (C 16 H 34 ) films on a crystalline substrate Au(001) In the simulation, the alkane system is described via intramolecular and intermolecular interactions between pseudoatoms (CH 2 and terminal CH 3 segments) and the ....

....Section 3.1.1 for example, it is necessary to synchronize the parameter modifications with the phases of the application to ensure that steering does not invalidate the computations. In some cases the design of the application makes this easy. The load balancing of the MD application described in [EGSM94] was facilitated because mechanisms were in place to handle molecules moving from domain to domain. These worked without modification when the domain boundaries themselves moved. In other cases it is clear that desired manipulations cannot be carried out without the direct cooperation of the ....

[Article contains additional citation context not shown here]

Greg Eisenhauer, Weiming Gu, Karsten Schwan, and Niru Mallavarupu. Falcon -- toward interactive parallel programs: The on-line steering of a molecular dynamics application. In Proceedings of The Third International Symposium on High-Performance Distributed Computing (HPDC-3), pages 26--34, San Francisco, CA, August 1994.

No context found.

Greg Eisenhauer, Weiming Gu, Karsten Schwan, and Niru Mallavarupu. Falcon - Toward Interactive Parallel Programs: The On-line Steering of a Molecular Dynamics Application. In Proceedings of the Third International Symposium on HighPerformance Distributed Computing, pages 26-34, August 1994.

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