AYDT, R. The pablo self-defining data format . Rapport Technique, Departement of Computer Science at the University of Illinois, Urbana-Champaign.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of processors. Following execution, performance data from each processor is integrated, additional statistics are computed, and the resulting metrics are correlated with application source code, creating a performance file that is represented via the Pablo self describing data for mat (SDDF) [11, 3]. This file is the specification used by SvPablo s browser to display application source code and correlated performance metrics. Use of the Pablo SDDF meta format has enabled us to develop a user interface that is both portable and language independent. Moreover, because all performance metrics ....

....for both instrumenta tion points and performance metrics. Only with this separation can one readily add new metrics and support new languages, compilers, and architectures without requiring extensive modifications to the user interface code. In SvPablo, the Pablo self defining data format (SDDF) [11, 3] provides this separation. SDDF defines data streams that consist of a group of record descriptors and record instances. Much as structure declarations in the C programming language specify templates for storage 10 allocation, SDDF descriptors define the structure for record instances. The data ....

AYDT, R. The Pablo Self-Defining Data Format. Tech. rep., Department of Computer Science at the University of Illinois at Urbana-Champaign, April 1994.

....library for PVM, and on the development of a single user interface to control a set of traditional sequential debuggers running over a PVM based application. 1 Monitoring support Concerning the specification of the trace file for a real execution, and according to the agreed SDDF based format[1], we have adapted our basic proposal in order to include suggestions from other partners. The new description, with the corresponding fields is given in annex A. Concerning the design of a distributed monitoring architecture supporting trace file generation, we have adopted a hierarchical ....

Ruth A. Aydt, The Pablo Self-Defining Data Format, Dep. Comp. Sc. - University of Illinois, 1994. 57

....transportation. Each monitoring instance may be responsible for a set of application processes, typically on a same physical machine, but possibly on several machines, depending on the system configuration. The generated event trace is based on the SDDF format, in order to ease tool integration [10]. More details about the monitoring tool can be found in [29] 30] 31] 32] The architecture that supports kernel and debugging functionalities has the following characteristics: ffl Dynamic attachment and deattachment of debugger instances to already running processes. ffl Control of remote ....

R. Aydt. The Pablo self-defining data format. Department of Computer Science, Univ. of Illinois, Apr 1994.

....Center has similar features. MAD [MAD99] from the GUP Linz in Austria provides event oriented tracing and analysis. Paradyn [MIL95] features performance tracing down to the procedure and statement level through dynamic instrumentation. Other tools propose visualization techniques, such as Pablo [AYD96] and Paragraph (part of the AIMS package) Paradyn, Upshot or even PARADE [PAR99] from Georgia Tech that also provide advanced navigation tools. However, as far as we know none of them address the issue of storing, and later accessing, the accumulated performance data. They may use proprietary ....

....make such a simplistic database quite impractical for a large number of measured experiments. There are ways to improve this solution. n A richer file format with embedded structure description can improve data availability. Possible candidates are the Self Defining Data Format (SDDF) [AYD96] or open standards such as SGML or its subset XML. n A simple Web site can offer an easy solution to the query issue, by hosting all the data in a standard format (most likely HTML or XML) The site then runs a search engine (e.g. Harvest [HAR96] which supports SGML, HTML and RTF) to index and ....

Ruth A. Aydt, "The Pablo Self-Defining Data Format", http://www.pablo.cs.uiuc.edu, 1996.

....performance measurement. Some of the most well known include: VAMPIR [VAM99] from the German company Pallas GmbH, for MPI programs, AIMS [YAN96] from NASA Ames Research Center, Paradyn [MIL95] that features performance tracing down to the statement level through dynamic instrumentation, and Pablo [AYD96] which is known for its visualization and self defining format, among other features. These tools focus much more extensively than our toolset on instrumenting codes to collect and compute performance measurement metrics and on locating computational bottlenecks. These and other tools also ....

Ruth A. Aydt, "The Pablo Self-Defining Data Format",http://www.pablo.cs.uiuc.edu.

....optimization mechanism, jitter control threshold, scaling technique, sphering technique, and a host of other parameters that configure the numerical analysis and optimization within PP. Our experimental infrastructure consists of a filter that accepts performance data in Pablo SDDF record format [1]. Each SDDF record contains a processor ID, a timestamp, and numerous metric value pairs. The filter reads these SDDF records directly from a file or from the Autopilot system [17] at runtime. The filter, then, 8 periodically emits a SDDF record that describes the relative importance of each ....

R. Aydt, "The Pablo Self-Defining Data Format," University of Illinois at UrbanaChampaign, Department of Computer Science, Urbana, IL, USA 1997.

....of PDRS Figure 2.1 depicts the design framework of PDRS. The technical approaches used to develop these components are discussed below section by section. 2. 1 Trace Data Module This module is in charge of collecting original performance data of parallel programs, and stores them with SDDF [1]. The large volume of data involved in parallel computations requires that instrumentation to collect the data selectively and intelligently. One way to collect data of a parallel program is to instrument the program executable so that when the program runs, it generates the desired information. ....

R. Aydt, The Pablo Self-Defining Data Format, Department of Computer Science, University of Illinois, April 1995, ftp://bugle.cs.uiuc.edu/pub/Release/Documentation/SDDF.ps.

....the Rice Fortran D compiler also creates a file that describes each program loop, procedure call and communication call in the generated code and its relation to the original Fortran D source code. This file, comprising static information about the program, is written in the Pablo SDDF format [5], and contains one record for each occurrence of those primitives (loop, procedure call and communication call) in the SPMD code. As an example, a message passing record specifies the message length (if known at compile time) the identity of the destination processor(s) the scalar or array ....

Ruth A. Aydt. The Pablo Self-Defining Data Format. University of Illinois at UrbanaChampaign, July 1994.

....using Pablo [30] a comprehensive performance analysis environment developed at the University of Illinois at Urbana Champaign. The Pablo environment includes instrumentation libraries that can record timestamped event traces, a performance data meta format called Self Defining Data Format (SDDF) [2] and associated library, a parser for generating instrumented SPMD source code, and tools for analyzing and graphically displaying the traces. Pablo also includes a special set of tools for capturing and analyzing I O performance data [3] The I O extension to the Pablo trace library provides the ....

....trace data across all processors, which can often reveal interesting patterns and peculiarities in the performance of the code. Upshot expects the logfiles to be in a specific format described in [16] The traces created by Pablo are in a different format called Self Defining Data Format (SDDF) [2]. Hence, we had to first convert Pablo trace files to the format required by Upshot. This conversion was trivial to do using Perl. 4.2 Machine Specifications The instrumented code was run on the IBM SP at Argonne National Laboratory and the Intel Paragon at Caltech, which are the two testbeds ....

R. Aydt. The Pablo Self-Defining Data Format. Technical report, Dept. of Computer Science, University of Illinois at Urbana-Champaign, March 1992.

....Both contain extensive support for data description and both address portability, though not to the extent we require. They are also oriented towards long term data storage and data exchange and so do not directly address issues of stream interpretation. The Pablo Self Defining Data Format (SDDF)[1] is a meta format designed to support monitoring trace files with requirements similar to our own. However SDDF s presumption that the sizes of basic data types do not change causes portability problems. While SDDF has a binary representation, its ASCII representation must be used for true ....

Ruth A. Aydt. The Pablo Self-Defining Data Format. Picasso Research Group, Department of Computer Science, University of Illinos at Urbana-Champaign, April 1994.

....a graphical language. This language describes the processing of the trace file through a graph where each node represents a specific operation or visualization to be applied to the input data coming from ancestors in the graph. Pablo reads trace files using the Self Defining Data Format (SDDF) [2] format which is a trace description language. Despite its programmability, we found Pablo difficult to use since the processing of our trace files require often complex computations which cannot be described easily with its graphical language. For these reasons, we decided to build a new tool, ....

R. A. Aydt. The Pablo Self-Defining Data Format. Technical report, University of Illinois, 1994.

....be the default. Ideally, trace data formats can be standardized across platforms, or tools can use trace specifications. Some effort has been devoted to trace standardization, but it has not gone far [9] We note, however, that there is a small trend toward converting trace files into SDDF format [2] for interchangeable tool use. 2.2.3 Syntax Analysis and Program Instrumentation Tools such as performance monitors and execution analyzers are far more effective if their interface is at the program level. That is, performance is reported in terms of source program structure and constructs, and ....

Aydt, R. The Pablo Self-Defining Data Format, 1993.

....dynamic, graphical animation of the behavior of message passing parallel programs, as well as graphical summaries of their performance. A concept demo of JPVS was prepared for the ARPA PI meeting held in June 96. JPVS helps to visualize execution traces (in Self Defining Data Format SDDF [11]) generated from Fortran or C codes instrumented with Pablo trace collection (instrumentation) library calls (Figure 5) Pablo [12] of the University of Illinois at Urbana Champaign is a well recognized performance instrumentation and analysis environment designed to organize and visualize ....

Aydt, R. A., "The Pablo Self-Defining Data Format," Technical Report, Department of Computer Science, University of Illinois, Sept. 1996). (At URL http://bugle.cs.uiuc.edu/Projects/ Pablo/documents.html ).

....Both contain extensive support for data description and both address portability, though not to the extent we require. They are also oriented towards long term data storage and data exchange and so do not directly address issues of stream interpretation. The Pablo Self Defining Data Format (SDDF)[1] is a meta format designed to support monitoring trace files with requirements similar to our own. However SDDF s presumption that the sizes of basic data types do not change causes portability problems. While SDDF has a binary representation, its ASCII representation must be used for true ....

Ruth A. Aydt. The Pablo Self-Defining Data Format. Picasso Research Group, Department of Computer Science, University of Illinos at Urbana-Champaign, April 1994.

....analysis. As we have noted, on line analysis is useful in situations where Avatar is being used to provide real time adaptive control, as discussed in x4.7. The Pablo instrumentation environment generates output in a Self Describing Data Format (SDDF) also developed at the University of Illinois [5]. This is a self contained data metaformat made up of record definitions and record instances. The record definitions describe all data types, sizes, and names used by a SDDF record while the record instances contain actual data values collected by the trace library. Figure 4.7 shows a example of ....

Aydt, R. A. The Pablo Self-Defining Data Format. Tech. rep., University of Illinois at Urbana-Champaign, Department of Computer Science, Apr. 1994.

....contain a new and improved tracing facility which provides more flexible and efficient access to run time program information. This new tracing system will support a buffering mechanism to reduce intrusion, and a more flexible trace event definition scheme based on SDDF (Self Defining Data Format) [3]. This new scheme expedites the collection and analysis of execution histories, and allows for integration of custom trace events. The tracing instrumentation is built into the PVM library to avoid re compilation, and supports on the fly adjustments to each task s trace event mask, to control the ....

R. A. Aydt, "The Pablo Self-Defining Data Format," Technical Report: University of Illinois at Urbana Champaign, Department of Computer Science, February 1993.

....tool such as XPVM using the regular PVM message channels. XPVM receives and processes these trace event messages to produce a trace file which can then be read in to drive graphical or textual views. The trace files for both tracing facilities are written in SDDF (Self Defining Data Format) [1] as designed for the Pablo project [4] SDDF was chosen over the PICL format [5] for the ParaGraph system [2] because PICL was not sufficiently flexible to represent the trace information generated by PVM. SDDF allows arbitrary data structuring and also includes some semantic information. The ....

R. A. Aydt, "The Pablo Self-Defining Data Format," University of Illinois at UrbanaChampaign, Department of Computer Science, February 1993.

....4 Trace format The trace format was defined within the APPARC project. It consists of kernel events and language events. The format of the kernel events is common to MYOAN, MAX, and ASVM, whereas the language events are specific to SVM Fortran. The events are defined in the standard SDDF format [17]. This format was chosen since the structure of the individual records can easily be changed without complex changes to the I O components of the tools. Furthermore, differences in the data representation between the parallel system and the platform of the analysis tool are handled automatically ....

Ruth A. Aydt, The Pablo Self--Defining Data Format, Department of Computer Science, University of Illinois, Urbana, Illinois 61801, 1993

....system side of this interface, this specifies the information required from the monitoring of the parallel computer system (or simulation) Interfacing subsets of this information to the visualisation tool enables the display of this information to take place. The basis for the interface is SDDF [1], a self defining data format developed at Illinois. This provides a flexible file format which can be used to specify data record structures and data record instances. The format is compact, portable, scalable and extensible. Binary and ASCII forms of this exist as do libraries to interface ....

Aydt, R.A., The Pablo Self-Defining Data Format, Department of Computer Science, University of Illinois, Illinois, USA, April 11, 1994.

....data analysis graph. As performance data travels down the analysis pipeline, it is converted into the desired performance metrics and graphical displays. Pablo achieves application independence through the use of a self defining trace data format. The Pablo Self Defining Data Format (SDDF) (Aydt 1992) is a meta format which specifies both data record structure and data record instances. SDDF provides the capability for user defined trace records which can be recorded and analyzed. The meta format allows for the creation of trace event descriptions which accompany specific trace records, ....

Aydt, Ruth A. 1992. The Pablo self--defining data format.

....the Rice Fortran D compiler also creates a file that describes each program loop, procedure call and communication primitive in the generated code and its relation to the original Fortran D source code. This file, comprising static information about the program, is written in the Pablo SDDF format [2], and contains one record for each program event. As an example, a message passing record specifies the message length (if known at compile time) the identity of the destination processor(s) the scalar or array section carried by the message, and the location of the original Fortran D source ....

Aydt, R. A. The Pablo Self-Defining Data Format. Department of Computer Science, University of Illinois at Urbana-Champaign, Dec. 1994.

....data across heterogeneous platforms, and (3) dynamic connectivity between transient applications. The substrate consists of: 1) event types cataloged and translated with PBIO [2] which supports event format and translation across heterogeneous machines with improved 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 ....

R. A. Aydt. Pablo Self-Defining Data Format. Department of Computer Science, University of Illinois, 1304 West Springfield Avenue, Urbana, Illinois 61801, May 1993.

....of processors. Following execution, performance data from each processor is integrated, additional statistics are computed, and the resulting metrics are correlated with application source code, creating a performance file that is represented via the Pablo self describing data format (SDDF) [3]. This performance file is the specification used by the SvPablo browser to display application source code and correlated performance metrics. Use of the Pablo SDDF metaformat has enabled us to develop a user interface that is both portable and language independent. Moreover, because all ....

Aydt, R. The Pablo Self-Defining Data Format. Tech. rep., Department of Computer Science at the University of Illinois at Urbana-Champaign, April 1994.

....amount of data collected during the program monitoring to a minimum and to guarantee the least intrusive monitoring experiment [26] capable of collecting all relevant data needed for the performance analysis. The format of the trace files is consistent with the Self Defining Data Format (SDDF) [2]. By generating trace records in SDDF we can use data processing and visualization facilities provided by Pablo, and also be able to translate the data into other formats, e.g. PICL, and then use ParaGraph for data visualization. Porthos. This tool supports instrumentation of a variety of source ....

R.A.Aydt, "The Pablo Self-Defining Data Format," Technical Report, University of Illinois at Urbana-Champaign, Department of Computer Science, 41 pages, Apr. 1994.

....Events related to language features are specific to either Fortran S [17] or SVM Fortran, they are described in the two following chapters. The trace format makes the assumption that an invalidation based protocol is used for coherence operations. The events are defined in the standard SDDF format [15]. This format was chosen since the structure of the individual records can easily be changed without complex changes to the I O components of the tools. Furthermore, differences in the data representation between the parallel system and the platform of the analysis tool are handled automatically ....

Ruth A. Aydt, The Pablo Self--Defining Data Format, Department of Computer Science, University of Illinois, Urbana, Illinois 61801, 1993

....instrumented by the compiler. Performance browsing is accomplished by correlating the performance data gathered at runtime with the source code, performing statistical analyses, and creating a performance file. This performance file, stored in the Pablo Self Defining Data Format (SDDF) [2], is used as input to the SvPablo tool which presents the performance information as it relates to the original source code. The Pablo SDDF is a performance data description language that specifies both data record structures and data record instances. This data meta format provides the following ....

Aydt, R. The Pablo Self-Defining Data Format. Tech. rep., Department of Computer Science at the University of Illinois at Urbana-Champaign, April 1994.

....for which only the format of the header has been defined. This header contains both the syntactic and the semantic description of the event records contained in a trace file. A tool reads this header to find out what information is contained in the trace file and where to find it [PGUB92, RP91, Ayd93] ffl object oriented : Instead of defining a trace format standard or just a header, the measured traceformats are considered as a generic abstract data structure. Evaluation tools can only access the measured data via a uniform and standardized set of generic procedures. In order to be able to ....

....but does affect its storage in the trace file. The data analysis The second part of Pablo is a graphical interface (using Motif) to analyze the data contained in trace files. The link between the instrumentation and the analysis is a trace file following the self describing data format (SDDF) Ayd93] This meta format allows for changes in the instrumentation without modifying the analysis tool. Obviously no semantic can be added in the trace file, so every field is recognized by a syntactic analysis. For example [RAM 92] Trace generation date November 1, 1991 ; #1: ....

R. Aydt. The pablo self-defining data format. Department of Computer Science, University of Illinois at Urbana-Champaign, March 1993. available by ftp anonymous bugle.cs.uiuc.edu:pub/Release-1.1/Documentation/SDDF.ps.Z.

....the real estate in the main window can be appropriately allocated between those two views by grabbing on a button at the left of the trace control panel and dragging it up or down. This button is shown in Figure 14. The trace files for XPVM are generated in self defining data format, or SDDF [1], as designed for representing trace events in the Pablo system system [7] SDDF was chosen over the PICL format [8] as used by the widely known ParaGraph system [2] because PICL was not sufficiently flexible to represent all the trace information generated by PVM. SDDF is a de facto ....

R. A. Aydt, "The Pablo Self-Defining Data Format," University of Illinois at Urbana-Champaign, Department of Computer Science, February 1993.

....the tutorials described in An Informal Guide to Using Pablo [2] An Overview of the Pablo Performance Analysis Environment [14] provides a more detailed overview of the Pablo environment and the philosophy behind it. More information about the Pablo Self Defining Data Format can be found in [1]. Information about the instrumentation system can be found in the users guides [11, 17] Information about sonification can be found in [4, 5, 6, 7] Other good documentation about the Pablo environment can be found in [3, 8, 9, 10, 12, 13, 15, 16] In addition to the Pablo documents referenced, ....

Aydt, R. A. The Pablo Self-Defining Data Format. Tech. rep., University of Illinois at Urbana-Champaign, Department of Computer Science, July 1994.

No context found.

AYDT, R. The pablo self-defining data format . Rapport Technique, Departement of Computer Science at the University of Illinois, Urbana-Champaign.

No context found.

Ruth A. Aydt, The Pablo Self--Defining Data Format, Department of Computer Science, University of Illinois, Urbana, Illinois 61801, 1993

No context found.

Ruth A. Aydt. The pablo self-defining data format. Technical report, Department of Computer Science, University of Illinois UC, March 1992.

No context found.

Aydt, Ruth A., "The Pablo Self-Defining Data Format," Pablo Working Documents, Dept. of Computer Science, UIUC, March 1992.

No context found.

R.A. Aydt. The Pablo Self-Defining Data Format. Department of Computer Science, University of Illinois at Urbana-Champaign, July 1994. Retrieved from ftp://www-pablo.cs.uiuc.edu/pub/Pablo.Release. 5/Documentation/SDDF.ps.gz.

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