B. Ries, R. Anderson, W. Auld, D. Breazeal, K. Callaghan, E. Richards, and W. Smith. The Paragon performance monitoring environment. In Proc. Supercomputing '93, pages 850-859. IEEE Computer Society, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....relevant works, like for example xpvm for assisting in debugging PVM applications [10] or ParaGraph, a performance visualisation tool Figure 9. Visualisation of a ProActive application, executing on 2 LSF allocated clusters of PCs inter connected with 1Gbits sec links. for Paragon applications [13]. The aforementioned environments target preferably parallel computers or clusters. As IC2D plus ProActive can indifferently run on any support, we could execute it in on various kind of machines, including ones not on the same administrative domain but accessible through Globus. and also on ....

B. Ries, R. Anderson, W. Auld, D. Breazeal, K. Callaghan, E. Richards, and W. Smith. The Paragon performance monitoring environment. In Proc. Supercomputing '93, pages 850-859. IEEE Computer Society, 1993.

....and unexpected performance problems for larger numbers of processors. In the next section we will start with a short survey on existing performance measurement tools. 2 Related work The performance measurement tools for parallel programs that have the widest distribution are Paragraph [7] [13] and the tools belonging to the EXPRESS environment [11] These tools support two levels of system representation in detecting and tracking down performance bottlenecks. On the first level profile data for the complete program execution is presented. There it is possible to evaluate the system ....

B. Ries, R. Anderson, W. Auld, D. Breazeal, K. Callaghan, E. Richards, and W. Smith. The paragon performance monitoring environment. In Proceedings of SuperComputing '93, pages 850 -- 859, Portland, Oregon, November 1993. IEEE Computer Society Press, Los Alamitos, CA.

....characterizing the phenomena to be observed [Ferrari83] A hybrid monitor is a combination of software and hardware. Many examples of software monitors can be found in the literature (see e.g. Plattner81; Power83; Malony92] Examples of hardware monitors are 26 COMTEN and SPM [Ibbett78] Ries93] describes the Paragon performance monitoring environment that uses a hardware performance monitoring board. And examples of hybrid monitors are Diamond [Hughes80] and HMS [Hadsell83] Each class of monitors has its own advantages and disadvantages. Selecting an appropriate monitor involves ....

....that must be considered in designing tools that monitor parallel programs. There are many parallel programming models available on computing systems (see e.g. Andrews91] so tools should be highly flexible. In tuning applications system effects must be distinguished from application bottlenecks [Ries93] The primary motivation of building multiprocessor systems is to cost effectively improve system performance. As we have seen earlier, given a parallel architecture, how to effectively use the computing power is the main effort in programming. It is obviously not efficient to obtain only a ....

Ries, B., Anderson, R., Auld, W., Breazeal, D., Callaghan, K., Richards, E. and Smith, W., "The Paragon performance monitoring environment," Proceedings of the conference on Supercomputing'93, pp. 850-859, 1993.

....Paragon. Commands are sent to the distributed monitoring system, called Tools Application Monitor (TAM) TAM consists of a network of TAM processes arranged as a broadcast spanning tree with one TAM process (part of the IS) at each node. Reference: http: www.ssd.intel.com paragon.html and [26] Debugging VIZIR This debugger consists of an integrated set of commercial sequential debuggers. Its IS synchronizes and controls the activities of individual debuggers that run the concurrent processes. The IS also collects data from these processes to run multiple visualizations. Reference: ....

Ries, Bernhard, R. Anderson, D. Breazeal, K. Callaghan, E. Richards, and W. Smith, "The Paragon Performance Monitoring Environment," Proceedings of Supercomputing `93, Portland, Oregon, Nov. 15--19, 1993.

....a separa (a ) b) Fig. 16. Paradyn IS testing results related to (a) Paradyn daemon process, and (b) main Paradyn process. TABLE 4 IS SERVICES USED BY TOOLS TO SUPPORT A RANGE OF FUNCTIONS Functionality Representative Tools Description of Key Services Performance evaluation ParAide [26] ParAide is the integrated tool environment for the Intel Paragon. Commands are sent to the distributed monitoring system, called Tools Application Monitor (TAM) TAM consists of a network of TAM processes arranged as a broadcast spanning tree with one TAM process (part of the IS) at each ....

# B. Ries, R. Anderson, D. Breazeal, K. Callaghan, E. Richards, and W. Smith, "The Paragon Performance Monitoring Environment," Proc. Supercomputing `93, Portland, Ore., pp. 850--859, Nov. 1993.

....of monitoring overhead to reduce the latency between the time an event is generated and the time it is acted upon for the purpose of steering. Various modules and functions of the IS are specified by a low level sensor specification language and a higher level view specification language. ParAide [29] is the integrated performance monitoring environment for the Intel Paragon. Commands are sent to the distributed monitoring system, called Tools Application Monitor (TAM) TAM consists of a network of TAM processes arranged as a broadcast spanning tree with one TAM process at each node. This ....

Ries, Bernhard, R. Anderson, D. Breazeal, K. Callaghan, E. Richards, and W. Smith, "The Paragon Performance Monitoring Environment," Proceedings of Supercomputing `93, Portland, Oregon, Nov. 15--19, 1993.

....perturbation, and implementation complexity. By recording the detailed temporal and spatial patterns of software interactions, timestamped event tracing subsumes the functions of profiling, timing, and counting (i.e. one can generate profiles, counts, and timing summaries from event traces) [12, 7, 14]. The flexibility of event tracing is not without price. If the events of interest occur frequently or the size of the instrumented system is large, tracing can quickly generate prodigious volumes of data. Moreover, unless the overhead for event recording and extraction is low relative to the ....

....invasive measures like counts. Although throttling prevents generation of extremely large data volumes, it sacrifices a consistent view of application behavior by periodically substituting counts for the more detailed traces. Others have proposed hardware support for event capture and processing [6, 14]. These techniques lower the overhead for event capture by providing data collection and extraction mechanisms that can operate in parallel with an application. Dynamic statistical clustering is complementary to all of these approaches. Thus, it can be used to automatically identify ....

B. Ries, R. Anderson, W. Auld, D. Breazeal, K. Callaghan, E. Richards, and W. Smith, The Paragon Performance Monitoring Environment, in Proceedings of Supercomputing '93, Nov. 1993, pp. 850--859.

....2 ParAide: An Integrated Tool Environment The Paragon s ParAide is a state of the art tool environment with industrial strength quality. It comprises a set of tools, all with a graphical user interface, that accompany the user from the application creation over debugging to performance analysis [5], 1] Table 1 gives an overview about the functionality of the main ParAide tools. What the user wants to do ParAide Tool What the tool enables to do program development in its initial phase requires to create the program or e.g. to change an existing sequential program to its parallel version. ....

B. Ries, R. Anderson, D. Breazeal, K. Callaghan. E. Richards, W. Smith. "The Paragon Performance Monitoring Environment". Proceedings of Supercomputing 93, Portland, Oregon. 1993.

....TDAS TDAS Extended PICL Figure 3: Performance Debugger Architecture RELATED WORK There are two main existing approaches to performance debugging of data parallel applications containing data distribution directives. The first one is based on the use of general purpose MIMD performance debuggers [5, 8]. The second approach is based on the static analysis of data parallel programs [1] Performance debugging tools designed for general message passing environments often provide the user with feedback in terms of low level events and state transformations such as message exchanges between ....

B. Ries, R. Anderson, D. Breazeal, K. Callaghan, E. Richards, and W. Smith. The Paragon performance monitoring environment. In Proceedings of Supercomputing '93, pages 850--859, Portland, Oregon, November 1993.

....the same machine. From the plots in Figure 7 we see that the speedups achieved with the Fibroblast code are smaller than those for the Bacteria code. To further investigate this and to learn more about the behavior of the codes, we used the performance monitoring environment on the Intel Paragon [19] to instrument the C code produced by the ZPL compiler. In particular, we used Xipd [10] to instrument the program and collect traces, and ParaGraph [12] to visualize performance. We produced detailed diagrams showing parallelism profiles, processor utilization, communication overhead, ....

B. Ries, R. Anderson, W. Auld, K. Callaghan, E. Richards, and W. Smith. The Paragon Performance Monitoring Environment. In Supercomputing '94, pages 850--859, 1994.

....like ndb or ipd [5] are based on pure textual interfaces that are difficult to use for non computer scientists. However, debuggers using graphical interfaces, e.g. Prism [6] are becoming more common. There is also a number of graphical interfaces built on top of text based debuggers, e.g. xipd [7] or Panorama [8] These tools usually provide visualization of arrays and thread states to ease debugging of complex applications. However, these advanced features are only usable for data parallel or SIMD style programs. Current debuggers don t support applications that create or delete threads ....

B. Ries, R. Anderson, W. Auld, D. Breazeal, K. Callaghan, E. Richards, and W. Smith. The paragon performance monitoring environment. In Proc. of Supercomputing '93, pages 850--859, Portland, Or., November 1993. IEEE.

....We instrumented malloc to keep track of dynamically allocated memory. In the case of INTEL s Paragon message passing multiprocessor, we instrumented communication routines manually to collect communication traces. Furthermore, we used the performance monitoring environment on the INTEL Paragon [26]: Xipd [8] for automatic instrumentation and trace collection and Paragraph [15] for performance visualization. On the KSR 2, we used manual instrumentation only, because the pmon monitoring libraries failed to produce reliable data. 1 2 4 8 12 16 32 60 64 128 Number of Processors 0 2000 4000 ....

B. Ries, R. Anderson, W. Auld, K. Callaghan, E. Richards, and W. Smith. The Paragon Performance Monitoring Environment. In Supercomputing '94, pages 850--859, 1994.

....dynamic data structures as supported by the HPC language. 7 Related Work There are two main existing approaches to performance debugging of data parallel applications containing data distribution directives. The first one is based on the use of general purpose MIMD performance debuggers [8, 14, 16, 17]. The second approach is based on the static analysis of data parallel programs [1] Performance debugging tools designed for general message passing environments often provide the user with feedback in terms of low level events and state transformations such as message exchanges between ....

B. Ries, R. Anderson, D. Breazeal, K. Callaghan, E. Richards, and W. Smith. The Paragon performance monitoring environment. In Proceedings of Supercomputing '93, pages 850--859, Portland, Oregon, November 1993.

....data and control transfer protocol (TP) is used for IS related communications. A majority of existing monitors use operating system supported interprocess communication abstractions (such as sockets in Pablo [22] and Issos [21] pipes in Paradyn [19] and remote procedure calls in TAM [24]) to accomplish this purpose. Some monitors (such as Hewlett Packard s VIZIR [7] implement customized high level protocols, developed on top of operating system functions, to enhance the flexibility and portability of the instrumentation data transfer and control messaging mechanisms. 6 2.3 ....

.... different developers are referred to as heterogeneous tools by Hao et al. 7] An integrated environment may support off line tool usage, such as TAU [2] and ParaVision [20] homogeneous on line tool usage, such as Paradyn [19] or a combination of the two, such as SPI [1] VIZIR [7] and ParAide [24]. Malony [18] presents a classification of measurement based tools comprising four classes: profilebased (sampling) trace based, prediction based, and automated (dynamic, adaptive, or knowledge based management) These types of tools, among others, are typically found in integrated environments. ....

[Article contains additional citation context not shown here]

Ries, Bernhard, R. Anderson, D. Breazeal, K. Callaghan, E. Richards, and W. Smith, "The Paragon Performance Monitoring Environment," Proceedings of Supercomputing `93, Portland, Oregon, Nov. 15--19, 1993.

....data structures as supported by the HPC language. 4 Comparison to Related Work There are two main existing approaches to performance debugging of data parallel applications containing data distribution directives. The first one is based on the use of generalpurpose MIMD performance debuggers [4, 7, 9, 10]. The second approach is based on the static analysis of data parallel programs [1] Performance debugging tools designed for general message passing environments often provide the user with feedback in terms of low level events and state transformations such as message exchanges between ....

B. Ries, R. Anderson, D. Breazeal, K. Callaghan, E. Richards, and W. Smith. The Paragon performance monitoring environment. In Proceedings of Supercomputing '93, pages 850--859, Portland, Oregon, November 1993.

....use binary re writing to insert instrumentation into an object file after it has been compiled and assembled. These systems require data collection decisions to be made prior to program execution. One system that defers instrumentation until the program has started to execute is the TAM facility[10] provided by Intel for the Paragon. TAM uses a static set of performance instrumentation profiles (i.e. prof style sampling, or full event tracing) to insert instrumentation into a program after it has been loaded into memory but prior to execution. Their method of inserting instrumentation is ....

B. Ries, R. Anderson, W. Auld, D. Breazeal, K. Callaghan, E. Richards and W. Smith, "The Paragon Performance Monitoring Environment", Supercomputing'93, Portland, OR, Nov 15-19, 1993, pp. 850-859.

....of processing the self describing data format and a graphical programming model that allows users to interactively connect and configure a data analysis graph. This Pablo software infrastructure has been used as a basis for a portion of Intel s performance analysis tools on the Intel Paragon XP S [19] and has been integrated with data parallel compilers to study the performance of Fortran D and HPF codes [1] For additional details on the Pablo software design philosophy and our experiences, see [15, 14, 16] 2.2 Pablo Software Extensibility Given the rapid changes in hardware platforms and ....

Ries, B., Anderson, R., Auld, W., Breazeal, D., Callaghan, K., Richards, E., and Smith, W. The Paragon Performance Monitoring Environment. In Proceedings of Supercomputing '93 (Nov. 1993), pp. 850--859.

....the data volume in bytes. 6 Unless the code is large and complex, with many types of messages generated by multiple computation phases, most users have a mental model of the expected interprocessor communication (e.g. ring or nearest neighbor on a grid) 6 ParaGraph [6] Intel s ParAide [12, 22] and many other tools contain similar displays. and can associate observed patterns with source code locations. However, if there are many types of messages or the same message type is generated from many source code locations, the mapping of performance data to code is more complex, and a ....

Ries, B., Anderson, R., Auld, W., Breazeal, D., Callaghan, K., Richards, E., and Smith, W. The Paragon Performance Monitoring Environment. In Proceedings of Supercomputing '93 (Nov. 1993), pp. 850--859.

....to collect the data. Pablo[75] and AIMS[101] 12 use source to source translation. The techniques described so far all require the instrumentation to be inserted prior to program execution. One system that defers instrumentation until the program has started to execute is the TAM facility[78] provided by Intel for the Paragon. TAM uses a fixed set of performance instrumentation profiles (i.e. prof style sampling, or full event tracing) to insert instrumentation into a program after it has been loaded into memory but prior to execution. TAM provides a way to instrument a program at ....

B. Ries, R. Anderson, W. Auld, D. Breazeal, K. Callaghan, E. Richards and W. Smith, "The Paragon Performance Monitoring Environment", Supercomputing'93, Portland, OR, Nov 15-19, 1993, pp. 850-859.

....monitoring systems largely avoid these data collection problems [3] but they are expensive and non portable, providing only low level system information without much flexibility. Software performance instrumentation is therefore generally chosen in preference to (or occasionally to complement [4]) a hardware solution, while recognizing the inherent problem of its intrusiveness. Analysis of the perturbations introduced, and schemes for ameliorating their impact have also been studied [5] While instrumentation can be carefully managed, and associated costs minimized, intrusion cannot be ....

B. Ries, R. Anderson, W. Auld, D. Breazeal, K. Callaghan, E. Richards, and W. D. Smith, "The Paragon performance monitoring environment," in Proc. Supercomputing '93 (Portland), pp. 850--859, IEEE, Nov. 1993.

....monitor to operate. These types of systems are not nearly as expensive as full hardware implementations but are slightly more intrusive. The MSPARC performance monitor [Harde92] the HMON Environment [Dodd92] the Multikron hardware [Harde95] and the Paragon Performance Monitoring Environment [Ries93] are based on the hybrid concept. Problems that need to be examined carefully when using hardware assistance for tracing are: w What kind of information will this provide the user w Is it adequate w Is it useful 15 w Will there be a need for higher level information that cannot be provided ....

....errors that it not hide these timing errors [Dodd92] MTOOL [Goldb91] attempts to reduce or eliminate problems encountered from perturbation by using minimum block counting techniques to reduce the time spent executing instrumentation instructions. The Paragon Performance Monitoring Environment [Ries93] allows for the specification of time ranges and processors for which trace data is to be collected. Specifying that only pertinent information be recorded reduces the amount of data that is generated and reduces the perturbation. Reducing the amount of data being recorded does not, however, allow ....

[Article contains additional citation context not shown here]

Bernhard Ries, R Anderson, W. Auld, D. Breazeal, K. Callaghan, E. Richards, and W. Smith, The Paragon Performance Monitoring Environment,Parallel Computation; Second International ACPC Conference, Springer-Verlag, 1993, pp. 233-248.

....3. Because the goal of data parallel languages is to insulate software developers from the idiosyncrasies of message passing, performance tuning should not require them to understand the details of the compiler generated code. Unfortunately, current performance instrumentation and analysis tools [9, 8, 6, 11] for distributed memory parallel systems only capture and present performance data from the generated Fortran 77 code of Figure 3. For example, Pablo performance displays for this program (shown in Figure 4) clearly show the performance characteristics of the message passing program, including the ....

Ries, B., Anderson, R., Auld, W., Breazeal, D., Callaghan, K., Richards, E., and Smith, W. The Paragon Performance Monitoring Environment. In Proceedings of Supercomputing '93 (Nov. 1993), Association for Computing Machinery, pp. 850--859.

....the goal of data parallel languages is to insulate software developers from the idiosyncrasies of message passing, performance tuning should not require them to understand the details of the compiler generated code. With the exception of MPP Apprentice [14] existing performance tools (e.g. [10, 9, 7, 12]) lack the ability to relate performance information from the compiler generated message passing code back to the source in the presence of substantial code restructuring by the compiler. Without access to compiler knowledge about the compilation process, tools are limited to presentation of ....

Ries, B., Anderson, R., Auld, W., Breazeal, D., Callaghan, K., Richards, E., and Smith, W. The Paragon Performance Monitoring Environment. In Proceedings of Supercomputing '93 (Nov. 1993), Association for Computing Machinery, pp. 850--859.

No context found.

Bernhard Ries, R Anderson, W. Auld, D. Breazeal, K. Callaghan, E. Richards, and W. Smith, The Paragon Performance Monitoring Environment, Parallel Computation; Second International ACPC Conference, Springer-Verlag, 1993, pp. 233-248.

No context found.

Bernhard Ries, R Anderson, W. Auld, D. Breazeal, K. Callaghan, E. Richards, and W. Smith, "The Paragon Performance Monitoring Environment," Parallel Computation; Second International ACPC Conference, Springer-Verlag, 1993, pp. 233-248.

No context found.

Bernhard Ries, Ray Anderson, W. Auld, Don Breazeal, Karla Callaghan, Eric Richards, and W. D. Smith. The Paragon performance monitoring environment. In Proceedings of Supercomputing'93 (Portland, OR, USA), pages 850--859. IEEE Computer Society Press, November 1993.

No context found.

B. Ries, R. Anderson, D. Breazal, et al. The Paragon Performance Monitoring Environment. In Proceedings SUPERCOMPUTING '93, Portland, OR, November 1993.

No context found.

B. Ries, R. Anderson, D. Breazal, et al. The Paragon Performance Monitoring Environment. In Proceedings SUPERCOMPUTING '93, Portland, OR, November 1993.

No context found.

B. Ries, R. Anderson, W. Auld, D. Breazeal, K. Callaghan, E. Richards, and W. Smith. The Paragon Performance Monitoring Environment. In Proceedings of Supercomputing '93, pages 850--859, Portland, Or., November 1993. IEEE.

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