D. M. Ogle, K. Schwan, and R. Snodgrass. Application-dependent dynamic monitoring of distributed and parallel systems. IEEE Transactions on Parallel and Distributed Systems, 4(7):762--778, July 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....devoted to traditional hard fault management issues in cluster environments. There have also been e#orts exploring performance issues in parallel program execution. These e#orts include evaluation of network e#ects [2, 7] performance analysis using application and kernel code instrumentation [4, 9, 11], performance prediction [8, 10] and program steering [3, 13] Additionally, adaptive techniques have been explored for predictive signaling and control in cluster environments for performance management [12] and in highly distributed networks for use in fault management [5] Our focus, however, ....

D. M. Ogle, K. Schwan, and R. Snodgrass. Application-dependent dynamic monitoring of distributed and parallel systems. IEEE Transactions on Parallel and Distributed Systems, 4(7):762--778, July 1993.

.... (eventdriven) By sampling, information about the execution state is collected synchronously (in a specific time rate) or asynchronously (through direct request of the monitoring system) By tracing, on the other hand, information is collected when an event of interest occurs in the system [11]. Tracing allows a better understanding and reasoning of the system behavior than sampling. However, tracing monitoring generates a much larger volume of data than sampling. In order to reduce this data volume problem, some researchers have been working on encoding techniques [12] A more common ....

D. M. Ogle, K. Schwan, and R. Snodgrass, "ApplicationDependent Dynamic Monitoring of Distributed and Parallel Systems", IEEE Trans. Parallel and Distributed Systems, vol. 4, n. 7, July 1993, pp. 762-778.

....monitoring agents) will be utilizing computing resources from the system in which they are running, thus affecting the system s behaviour. Therefore, a major concern with any monitoring system is its effect on the performance of the system. Related work in existing monitoring systems include [1, 3, 5, 6]. In [3] data is collected independent of its use. This results in the collection of large amounts of information that may not be used. On the other hand, the monitoring systems described in [1, 5, 6] allow for the formulation of directives that describe the data to be collected and the analysis ....

....effect on the performance of the system. Related work in existing monitoring systems include [1, 3, 5, 6] In [3] data is collected independent of its use. This results in the collection of large amounts of information that may not be used. On the other hand, the monitoring systems described in [1, 5, 6] allow for the formulation of directives that describe the data to be collected and the analysis to be done. However, in [5] the majority of the analysis is done in a single central process. The use of a central process means that there is the potential for the central process to become a ....

[Article contains additional citation context not shown here]

D. Ogle, K. Schwan, and R. Snodgrass. Application -dependent dynamic monitoring of distributed and parallel systems. IEEE Transactions on Parallel and Distributed Systems, 4(7):762--778, July 1993.

....The ERS is a HiFi library linked with the monitored application during compilation to facilitate event reporting. The main function of the instrumentation component is to facilitate the process of inserting the monitoring instruction (sensors) inside the program code. In many monitoring systems [17, 20], programmers write a considerable size of code for each generated event. This makes the instrumentation task tedious and error prone. In HiFi, users only insert the user sense that specifies just the event name (e.g. ReportEvent(WarningEvent) The instrumentation component then pre processes ....

D. Olge, K. Schwan, and R. Snodgrass. Application-Dependent Dynamic Monitoring of Distributed Systems. IEEE Transactions on Parallel and Distributed Systems, 21(4):593--622, December 1989.

....traffic) since resource availability and other system and user requirements are not taken into consideration. We use the term monitoring overhead to refer to the resource usage by monitoring. Researchers have recently begun to address these issues using different approaches. Earlier work (e.g. [4, 5, 6]) focusses on the control of the amount of data to be transferred from managed objects to management agents, by specifying queries. Queries are directives issued to management agents, specifying the managed objects to be monitored, the data to be collected and the analysis to be done on the ....

D. Ogle, K. Schwan, and R. Snodgrass. Applicationdependent dynamic monitoring of distributed and parallel systems. IEEE Transactions on Parallel and Distributed Systems, 4(7):762--778, July 1993.

....to do relational database queries [15] This is limited to SNMP and assumes that the agents are already started up. Our approach to defining the time semantics in QDL was influenced by the temporal query language research by Snodgrass and his colleagues [13] Other relevant work includes [5, 10]. In [10] the majority of the analysis is done in a single central process. In our work, analysis may be distributed by through management agents (provided management agents exist that can do the analysis) The Coordinator can direct that management agents collecting raw data send this raw data to ....

....relational database queries [15] This is limited to SNMP and assumes that the agents are already started up. Our approach to defining the time semantics in QDL was influenced by the temporal query language research by Snodgrass and his colleagues [13] Other relevant work includes [5, 10] In [10] the majority of the analysis is done in a single central process. In our work, analysis may be distributed by through management agents (provided management agents exist that can do the analysis) The Coordinator can direct that management agents collecting raw data send this raw data to the ....

David Ogle, Karsten Schwan, and Richard Snodgrass. Application-dependent Dynamic Monitoring of Distributed and Parallel Systems. IEEE Transactions on Parallel and Distributed Systems, 4(7):762--778, July 1993.

....on execution information regarding the states and behavior of application programs to provide useful feedback to the user. With increasing sophistication of parallel tool development technologies, runtime data collection and management activities are receiving more attention from tool developers [26]. Parallel tool developers are focusing on integrated parallel tool environments [37] and frameworks [32] performance evaluation of real time systems [2] and program steering [7] The overheads and perturbation effects associated with data collection and management are of critical importance in ....

....to actual measurements during instrumented program execution. SPI [2] ensures that the invasiveness of its IS is accountable. It measures the instrumentation load on nodes and links in each specified window of time to evaluate the degree of invasiveness relative to an application program. Falcon [26] is perhaps the only tool that supports a thorough evaluation of its instrumentation system. Perturbation to programs is measured under different conditions of tracing rates, event record lengths, and event buffer sizes. On the fly ordering of event records, which is needed for meaningful ....

Ogle, David M., Karsten Schwan, and Richard Snodgrass, "Application-Dependent Dynamic Monitoring of Distributed and Parallel Systems," IEEE Transactions on Parallel and Distributed Systems, 4(7), July 1993, pp. 762--778.

....authors present the straightforward approach of introducing the monitoring activity at the user selected points and [20] presents a hardware solution to non intrusive monitoring by providing architectural support. Our approach can enhance the effectiveness of monitoring systems such as the ISSOS [13]. In the next section we will present the models used to characterize the real time application program and the monitoring tasks. In section 3 we present timing analysis algorithms by which monitoring tasks are integrated into a real time application task. In section 4 we present extensions to ....

D.M. Ogle, K. Schwan, and R. Snodgrass, "Application-Dependent Dynamic Monitoring of Distributed and Parallel Systems," IEEE Transactions on Parallel and Distributed Systems, Vol. 4, No.7, July 1993.

....might have on application performance. More specifically, one proposed adaptation cost model describes adaptation overhead with an off line measured delay parameter [56] and assumes that the adaptation can occur anytime during application execution [3] Similarly, research in program monitoring [59, 52] has introduced explicit specifications of delay in the capture of monitoring and has developed methods for reducing such delay. A richer cost model is necessary for a real time system. The motivation is twofold. First, an application cannot always execute the adaptation procedure as soon as ....

D.M. Ogle, K. Schwan, and R. Snodgrass. Application-dependent Dynamic Monitoring of Distributed and Parallel Systems. IEEE Transactions on Parallel and Distributed Systems, July 1993.

....tasks. Increased flexibility is achieved through a declarative and comprehensible monitoring language, simple code instrumentation process, and automated monitoring agents administration. A number of monitoring approaches and systems for monitoring distributed systems have been proposed (e.g. [5, 6, 7, 8, 10, 11, 12, 13, 14]) Although some of these systems have various design goals and objectives, they are insufficient to support a scalable and fine grain event filtering mechanism for monitoring large scale distributed systems such as Internet based applications. In addition, the monitoring filtering tasks in such ....

D. Olge, K. Schwan, and R. Snodgrass. ApplicationDependent Dynamic Monitoring of Distributed Systems. IEEE Transactions on Parallel and Distributed Systems, 21(4):593--622, December 1989.

....without interfering with the running application, and specifying programmable management tasks that can automatically initiate new monitoring operations based on the events detection. A number of monitoring approaches and systems for monitoring distributed systems have been proposed (e.g. [4, 10, 11, 13, 16, 17, 18, 19, 21, 22, 24]. Although some of these systems have various design goals and objectives, they are insufficient to support a scalable and fine grain event filtering mechanism for monitoring large scale distributed systems such as Internet based applications. In addition, the monitoring filtering tasks in such ....

....The ERC is then used by the Event Reporting Stub (ERS) for constructing and reporting the event notification. The main function of the instrumentation component is to facilitate the process of inserting the monitoring instruction or sensors inside the program code. In many monitoring systems [17, 19], programmers write a considerable size of code for each generated event. This makes the instrumentation task tedious and error prone. In HiFi, users only inserts the user sense that just specifies the event name (e.g. ReportEvent(WarnEvent) The instrumentation component then pre processes the ....

D. Olge, K. Schwan, and R. Snodgrass. Application-Dependent Dynamic Monitoring of Distributed Systems. IEEE Transactions on Parallel and Distributed Systems, 21(4):593--622, December 1989.

....a program so that interesting dynamic data and events can be captured. The gathered information may support a wide spectrum of inspection tasks, including optimization[HE89, LSV 89, LBSV90] debugging[HC87] testing[Bat88] reverse engineering[RS93, MA96] or dynamic, on line reconfiguration[OSS93, SC95b] While instrumentation techniques have historically varied[KS93] the most widely used scheme is to insert event sensors at selected points in the program. Each event sensor monitors particular program events (for example, method calls) and reports them to, say, a trace recorder. Figure ....

David M. Ogle, Karsten Schwan, and Richard Snodgrass. ApplicationDependent Dynamic Monitoring of Distributed and Parallel Systems. IEEE Transactions on Parallel and Distributed Systems, Vol. 4, NO. 7, pages 762-- 778, July 1993.

....on execution information regarding the states and behavior of application programs to provide useful feedback to the user. With increasing sophistication of parallel tool development technologies, runtime data collection and management activities are receiving more attention from tool developers [21]. Parallel tool developers are focusing on integrated parallel tool environments [30] and frameworks [26] performance evaluation of real time systems [1] and program steering [6] The overheads and perturbation effects associated with data collection and management are of critical importance in ....

.... level (software, hardware, or hybrid) trigger mechanism (eventdriven or time driven, i.e. sampling) and data presentation and analysis mode (on line or off line) For a parallel or distributed system, a monitor is responsible for the collection and analysis of distributed program information [21, 27]. 2.2 Instrumentation System An instrumentation system (IS) is the part of a monitor that is concerned with the collection and management of performance information. The scope of an IS spans three modules or functions that are components of any software monitor for a parallel or distributed ....

[Article contains additional citation context not shown here]

Ogle, David M., Karsten Schwan, and Richard Snodgrass, "Application-Dependent Dynamic Monitoring of Distributed and Parallel Systems," IEEE Transactions on Parallel and Distributed Systems, 4(7), July 1993, pp. 762--778.

....component receives directives or queries from management applications and directly or indirectly from administrators or from other management services. The queries describe the data and information to be collected and the analysis to be done. Previous work in existing monitoring systems include [7, 22, 28]. In these monitoring systems the majority of the analysis is done in single central process or the information to be collected must be known a priori or a fixed set of queries is assumed. What is needed in an approach that allows for the evaluation of a dynamic set of queries and delegating ....

David Ogle, Karsten Schwan, and Richard Snodgrass. Application-dependent dynamic monitoring of distributed and parallel systems. IEEE Transactions on Parallel and Distributed Systems, 4(7):762--778, July 1993.

....to interleave them has been made. On the other hand Gerber and Hong s approach requires analyzing the timing behavior of the tasks resulting from the slicing transformation. Our approach for non intrusive monitoring of real time applications can enhance the effectiveness of existing systems (Ogle, Schwan, and Snodgrass, 1993, Tokuda, Kotera, and Mercer, 1988, Haban and Shin, 1989) In contrast to the non intrusive monitoring techniques developed by Tsai, Fang, and Chen (1990) which require specialized hardware support, our approach is based entirely upon software techniques. 6. Concluding Remarks We presented ....

Ogle, D.M., K. Schwan, and R. Snodgrass, "Application Dependent Dynamic Monitoring of Distributed and Parallel Systems," IEEE Transactions on Parallel and Distributed Systems, July, 1993, 4(7):762--778.

....monitoring systems. 1. Interactive monitoring systems Interactive monitoring systems permit users to interact with the monitoring system during the execution of the target application. Users can enable, disable, and reconfigure the monitoring system to suit their exploration of the application [19, 17, 9, 6]. These interactive monitoring systems complement other forms of monitoring that use post mortem analysis [25, 18, 16, 2, 3] but they share with such systems the need to analyze and display monitoring data [20, 11, 12, 4] The interactive monitoring systems # NASA financially supported Vetter ....

....systems, which include Paradyn, Quartz, and EEL [17, 2, 3] use modified compilers or executable editing to interrogate the application for performance relevant information. However, we do not employ these techniques because they are limited in the types of information they can provide to the user [19], especially when executables are highly optimized [7, 24] Alternatively, users of application specific monitoring systems annotate their application source code to supplement information available from these other monitoring methods. Our particular form of application specific monitoring, called ....

[Article contains additional citation context not shown here]

D. Ogle, K. Schwan, and R. Snodgrass. Applicationdependent dynamic monitoring of distributed and parallel systems. IEEE Trans. Parallel and Distributed Systems, 4(7):762--778, 1993.

....with a novel approach to selectively extracting data from data streams. Earlier work done by our group has established the benefits of encapsulating needs mismatch style computations into logical tasks that can be associated with data streams to maximize proximity or availability of resources [25, 5, 17, 21]. This paper extends these notions by providing the user with an intuitive relational model for thinking about needs mismatch computations and a prototype system for creating these computations and embedding them into a data stream. Application of the work first to safety critical systems[24] ....

....end to end latency. We conclude with a summary of contributions and future work in Section 9. 2 Related Research One familiar application of a valid time database to parallel and distributed computing is the use of temporal SQL for performance analysis of distributed and parallel computations [30, 21, 18]. Queries are issued post mortem against the performance data[30] Our goal, on the other hand, is analysis at run time, which puts greater demands on efficiency. Further, by pushing analysis into the data stream[21] one can reduce the amount of data that must be ultimately stored, and the period ....

[Article contains additional citation context not shown here]

David Ogle, Karsten Schwan, and Richard Snodgrass. Application-dependent dynamic monitoring of distributed and parallel systems. IEEE Transactions on Parallel and Distributed Systems, 4(7):762--778, July 1993.

....by others in [3] 4] and [8] Our work with dQUOB is in adapting database queries to operate over streaming data instead of database tables. Viewing data streams as data sources over which relational queries can be specified has been explored in the past in the context of performance monitoring [19, 12], but it suffered limitations in the ability to keep up. Our contribution is to replace all traces of a database with temporary buffers to improve performance. 1 The work further contributes adaptivity to query processing, under the hypothesis that more efficient, optimal queries can be achieved ....

David Ogle, Karsten Schwan, and Richard Snodgrass. Application-dependent dynamic monitoring of distributed and parallel systems. IEEE Transactions on Parallel and Distributed Systems, 4(7):762--778, July 1993.

....applications able to interact at human speeds via user interfaces, such as the application presented in this paper. Regardless of the speed and nature of interaction, program steering is based on the on line capture of information defining current program and configuration state[Bru91, MW91, OSS93] and it assumes that human users or algorithms inspect such information and manipulate it to make steering decisions. This paper focuses on the opportunities and costs of on line steering as applied to one substantial parallel application, a molecular dynamics simulation (MD) used by physicists ....

D.M. Ogle, K. Schwan, and R. Snodgrass. Application-dependent dynamic monitoring of distributed and parallel systems. IEEE Transactions on Parallel and Distributed Systems, 4(7):762--778, July 1993.

.... computing continues to remain non interactive [McCormick88] end users and program developers alike will not capitalize on new techniques for interactive data visualization and program animation [Jablonowski93,Stasko90] remote and collaborative work, interactive debugging and monitoring [Gu95,Ogle93], and on line program adaptation [Mukherjee93] For example, when run in batch mode , erroneous or uninteresting results produced by large scale scientific or engineering simulations are not apparent until after the computations complete, sometimes days or weeks after program initiation. Program ....

....and places the record into a buffer destined for the monitoring system. Because sense is executed within the control flow of the application, sense is synchronous. The steering server can enable or disable Sense for each particular steering object. Both probes and sensors are investigated in [Ogle93] as part of an application specific monitoring system. In Progress, these mechanisms monitor steering objects instead of language specific application components. Actuate performs a modification on the steering object. Actuate is a new steering mechanism that is analogous to the sense mechanism ....

David M. Ogle, Karsten Schwan, Richard Snodgrass. "Application Dependent Dynamic Monitoring of Distributed and Parallel Systems." IEEE Transactions on Parallel and Distributed Systems, 4(7):762-778, July 1993.

....action execution. The first three steps are common to event based monitoring. The last four steps are specific to the proposed steering model. 2. 1 Monitoring Monitoring is the extraction of dynamic information from a computational system[14] This model uses an event based approach to monitoring[15, 16, 17, 18], where each primitive event contains 4 basic components[16] represented by the tuple event class, processor ID, timestamp, state . Primitive events are immutable. Event class is an attribute that describes the type of the event. Event classes differ in the set of attributes they capture. In MD, ....

....however, we list them here to solidify the model s usefulness. 4.1 Monitoring Mechanisms The results of any monitoring mechanisms are events as described earlier. Two possible types of monitoring mechanisms produce events: sensors and probes. Both of these devices are described elsewhere [17]. Sensors. Sensors are instrumentation points placed throughout the application s source code; therefore, sensors are synchronous mechanisms with respect to the application. Each sensor captures information about the state of the application or a system resource as an event and forwards these ....

D. M. Ogle, K. Schwan, and R. Snodgrass, "Application-dependent dynamic monitoring of distributed and parallel systems," IEEE Trans. on Parallel and Distributed Systems, vol. 4, pp. 762-- 78, 1993.

.... activities observed by the monitor are specified by a high level query language called TQuel (considerably refined in [30] Based, in part, on the relational approach originated by Snodgrass, Ogle et al. explore application dependent and on line monitoring of parallel and distributed applications [18]. Kilpatrick and Schwan [11] further refine the language based approach for specifying application dependent monitoring requests. Kilpatrick also explores the idea of integrating all components of a parallel programming environment using the uniform Entity Relationship information model. Prior ....

David Ogle, Karsten Schwan, and Richard Snodgrass. Application-dependent dynamic monitoring of distributed and parallel systems. IEEE Transactions on Parallel and Distributed Systems, 4(7):762--778, July 1993.

.... activities observed by the monitor are specified by a high level query language called TQuel (considerably refined in [22] Based, in part, on the relational approach originated by Snodgrass, Ogle et al. explore application dependent and on line monitoring of parallel and distributed applications[16]. Kilpatrick and Schwan[9] further refine the language based approach for specifying application dependent monitoring requests. Kilpatrick also explores the idea of integrating all components of a parallel programming environment using the uniform Entity Relationship information model. As ....

David Ogle, Karsten Schwan, and Richard Snodgrass. Application-dependent dynamic monitoring of distributed and parallel systems. IEEE Transactions on Parallel and Distributed Systems, 4(7):762--778, July 1993.

....able to interact at human speeds via user interfaces, such as the application presented in this paper. Regardless of the speed and nature of interaction, program steering is based on the on line capture of information defining current program and configuration state[Bru91, MW91, SRVO87, OSS93] and it assumes that human users or algorithms inspect such information and manipulate it to make steering decisions. This paper focuses on the opportunities and costs of on line steering with one substantial parallel application, a molecular dynamics simulation (MD) used by physicists for ....

....slabs of the material being simulated. One additional thread is collecting and computing global state like total energy and performing visualization processing (i.e. displaying current particle and molecule positions) The Falcon tool uses a second additional thread, called a local monitor thread[OSS93] which collects all monitoring outputs from the running program, filters them as much as possible, and then uses well defined interfaces to either store monitoring information or forward it to any interested party, which may include: 1) on line steering algorithms, 2) display tools, or (3) ....

D.M. Ogle, K. Schwan, and R. Snodgrass. Application-dependent dynamic monitoring of distributed and parallel systems. IEEE Transactions on Parallel and Distributed Systems, 4(7):762--778, July 1993.

....program may be steered or adapted monitoring information captured with Falcon may be attached to arbitrary user provided analysis code, graphical views for output or program steering, and adaptation algorithms. Analyses may employ statistical methods, boolean operators like those described in [41], or simply reorder the events being received. Graphical views may be displayed with multiple media or systems, currently including X windows, Motif, and the SGI Explorer environment. For brevity, this paper does not elaborate on two additional aspects of Falcon, which are (1) the support of ....

....interact with the runtime system to obtain the partially processed monitoring 2 Stub insertion may not be trivial. Future work should address compiler and user interface support for this task, as well as additional functionality in the stub compiler to generate alternative stub implementations[41]. information. Further analysis of the trace information is performed before it is displayed to end users or used in steering algorithms. Trace information can also be stored in trace files for postmortem analysis. Once steering decisions are made by the end user or a steering algorithm, changes ....

[Article contains additional citation context not shown here]

D. M. Ogle, K. Schwan, and R. Snodgrass. Application-dependent dynamic monitoring of distributed and parallel systems. IEEE Transactions on Parallel and Distributed Systems, 4(7):762--778, July 1993.

First 50 documents

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