Tracing Summit Presentations 2008

= Day 1 - Tuesday January 29, 2008 =

9h00-10h20 Michel Dagenais, Gabriel Matni, Pierre-Marc Fournier, Robert Roy  Survey of the best available tools and techniques for system performance tracing and analysis  ([[Media:TracingReview_MichelDagenais_MontrealJan2008.pdf|part1]], [[Media:TracingReview_GabrielMatni_MontrealJan2008.pdf|part2]], [[Media:TracingReview_PierreMarcFournier_MontrealJan2008.pdf|part3]], [[Media:TracingReview_RobertRoy_MontrealJan2008.pdf|part4]]) The talk will describe the different types of tools and techniques used to extract, analyze and visualize system execution and performance data. The survey will cover source level and binary instrumentation (e.g. Dtrace, SystemTap), as well as sampling (e.g. gprof, OProfile) for extracting data. Different visualization and analysis tools and frameworks will be described, including tools from the Eclipse framework, QNX and Wind River. As will be described by Robert Roy, for large parallel computing applications, the ultimate challenge of parallel performance analysis tools is to help in finding bottlenecks and synchronization problems and further optimizing the code. He will focus on the use of performance analysis tools for developing parallel solvers (inputs, ease of use, scalability).

10h30 – 10h50 Frank C. Eigler / RedHat  SystemTap  ([[Media:SystemTap_FrankEigler_MontrealJan2008.pdf|slides]]) Description of the SystemTap free software infrastructure to simplify the gathering of information about running Linux systems. SystemTap eliminates the need for the developer to go through the tedious and disruptive instrument, recompile, install, and reboot sequence that may be otherwise required to collect data.

11h00-11h20 Andrew Cagney / RedHat  Frysk; and tracing from userland  ([[Media:Frysk_AndrewCagney_MontrealJan2008.pdf|slides]]) The Frysk group are implementing a framework of tools for, tracing, monitoring and debugging large distributed applications. After a brief overview of Frysk and its architecture; this presentation will examine some of the challenges with us taking a largely user-land approach. Ptrace and utrace, and DWARF will each be discussed.

11h30 – 12h20 Robert Wisniewski / IBM  Performance analysis and debugging at IBM  ([[Media:PerformanceAnalysisIBM_RobertWisniewski_MontrealJan2008.pdf|slides]]) Discussion of the performance analysis and debugging he has done at IBM. He will describe the challenges presented by both external and internal IBM clients, typically in the area of HPC and multi-core servers. He will then describe the tools he has worked on to address those challenges.

13h30-14h20 Eugene Chan / IBM - Eclipse Test & Performance Tools Platform  Extending Eclipse Test and Performance Tools Platform (TPTP)  ([[Media:ExtendingTPTP_EugeneChan_MontrealJan2008.pdf|slides]]) Overview of the TPTP architecture following by a quick look at the profiling features of TPTP. Finally, we will focus on UI extensions that are available to support non-Java context.

14h30- 15h20 Steven Shaw / IBM Rational  IBM Rational Systems Developer Extensibility with consideration of UML Debug and Trace integrations  ([[Media:UMLTracing_StevenShaw_MontrealJan2008.pdf|slides]]) This presentation will provide a brief overview of the Rational Systems Developer tooling and then drive into the different extensibility features that it offers. The architecture of the product with respect to the open source components that it leverages will be explored as well as the relevant extensibility of those components. For example, we will briefly look at the extensibility mechanisms for Eclipse platform, EMF (Meta-model framework), UML2 (concrete meta-model built using EMF), GEF (Graphical Editing Framework), GMF (Graphical / Generative Modeling Framework), RSD (Rational Systems Developer). Finally we will create / examine some example plug-ins / plug-lets that utilize the extensibility with consideration of typical integrations that would be important for debugging a UML system.

15h30-15h50 Elena Zannoni / Oracle  Tracing at Oracle  ([[Media:OracleTracing_ElenaZannoni_MontrealJan2008.pdf|slides]]) Elena will talk about the experience of the Oracle DataBase team with tracing and the problems they face.

16h00-16h20 Mathieu Desnoyers / Ecole Polytechnique  Tracing the Google platform  ([[Media:TracingGooglePlatform_MathieuDesnoyers_MontrealJan2008.pdf|slides]]) Mathieu will describe his experience in tracing while working for the platform team at Google. He will describe a number of typical hard to diagnose problems encountered on extremely large clusters and how they were solved using tracing tools. The specific challenges and expectations for very large high performance online server clusters will be exposed.

16h30-16h50 Steve Furr / Freescale  Multicore Enablement Challenges for Semiconductor Manufacturers  ([[Media:FreescaleMulticoreMonitoring_SteveFurr_MontrealJan2008.pdf|slides]]) This presentation will highlight some of the challenges that face vendors as multicore processors become more prevalent. Multicore solutions increase the complexity of static and realtime debugging tools. The difficulties, motivations and objectives behind enabling multiple vendors of dynamic debug and profiling tools are examined from the perspective of a semiconductor manufacturer's perspective.

17h00-17h20 Stuart Fullmer / MontaVista  Tracetools with MontaVista Linux  (slides not provided) Short presentation regarding LTTng and TimeDoctor. The presentation will give a quick overview of the tools and our level of integration of the tools and where MontaVista sees itself with future offerings.

19h00 – 20h00 BOF (Birds Of a Feather) sessions
 * System Level Target Data Collection
 * Eclipse Tracing Monitoring Framework

= Day 2 - Wednesday January 30, 2008 =

9h00-9h50 Sven Lundblad / ENEA  Advanced Event - Action System and Flexible Profiling  ([[Media:ENEAEventActionSystem_SvenLundblad_MontrealJan2008.pdf|slides]]) Enea's advanced Event Action system. Use cases (system tracing and debugging), functionality (event types, filtering, different actions, event action states), intrusiveness, and controlling tools. Possibilities and challenges in multi core systems. The Event Action implementation in OSE and possible Linux support. Optima Profiling capabilities Advanced custom profiling, memory profiling and CPU usage profiling (in single core and SMP systems).

10h00 – 10h20 Henrik Thane / ZealCore  Monitoring, Information Fusion, Reverse Engineering and Replay Debugging  ([[Media:ZealCoreMonitoringReplay_HenrikThane_MontrealJan2008.pdf|slides]]) Discussion on different trace recording approaches including handling of multi-core systems, and most importantly what do you do with the recorded information? How can you fuse different trace logs, do reverse engineering and raise the level of Debugging? We will especially look into the problem on how to achieve forward and backward replay debugging on the UML model level (reanimated state-machines and sequence diagrams) based on reverse engineering of minimal target run-time recordings.

10h30-11h20 Beth Tibbitts / IBM Research  Eclipse Parallel Tools Platform & Performance Tools Framework  ([[Media:EclipseParallelToolsPlatform_BethTibbitts_MontrealJan2008.pdf|slides]]) The Eclipse Parallel Tools Platform (http://eclipse.org/ptp) aims to produce an open-source industry-strength platform that provides a highly integrated environment specifically designed for parallel application development. This includes a standard, portable parallel IDE that supports a wide range of parallel architectures and runtime systems, a scalable parallel debugger, support for the integration of a wide range of parallel tools, an environment that simplifies the end-user interaction with parallel systems. A performance tools framework is being designed to ease the integration of performance tools into Eclipse and PTP. The University of Oregon has developed Eclipse plugins for PTP for its TAU (Tuning and Analysis Utilities) and we are working to generalize and make these features and others available to ease integration of other tools, including performance tools with PTP and EclipseParallel computing tracing with Eclipse Parallel Tools Platform Performance Analysis Framework / Tuning and Analysis Utilities.

12h30-13h20 Andrew McDermott / WindRiver  Developing OS-agnostic visualization tools  ([[Media:SystemViewer_AndrewMcDermott_MontrealJan2008.pdf|slides]]) Today it is desirable to use a common [visualization] tool for diagnosing system behaviour. However, it is more common to find different tools, different user interfaces, different data formats and different levels of overall tool capability on a [operating] system-by-system basis. Additionally, the mechanics of querying the data in an ad-hoc manner outside the boundaries of the tool are typically programming language specific or non-existent. Wind River's System Viewer, previously known as WindView, is a sophisticated visualization tool that enables developers to view the dynamic operation of their system. System Viewer has evolved to address the deficiencies listed above to be an OS-agnostic visualization tool (e.g., VxWorks and LTT), providing a separate and SQL-query able data engine that can accommodate new systems.

13h30-14h20 Felix Burton / WindRiver  Target Communication Framework  ([[Media:TargetCommunicationFramework_FelixBurton_MontrealJan2008.pdf|slides]]) Today almost every device software development tool on the market has its own method of communicating with the target system. Communication methods often conflict with each other, requiring individual setup, configuration, maintenance, and imposes different limitations for different tools. TCF is designed to establish a common communication mechanism between development tools and embedded devices for the purposes of debugging, performance monitoring, file system access, etc.

 Wind River Sensorpoint Technology  ([[Media:SensorPoint_FelixBurton_MontrealJan2008.pdf|slides]]) Increasingly complex systems are hard to debug using traditional techniques such as breakpoints and stepping or printf. Sensorpoints allow dynamic instrumentation of software in the lab as well as in the field with minimal impact on the system performance and system resources. This makes it ideal for diagnosing complex problems as well as creating patches to fix such issues in deployed devices. Sensorpoints are code fragments that can be dynamically “injected” into a binary image, without requiring a restart of the application or the device.

14h45-16h15 Michel Dagenais, Robert Charpentier, Dominique Toupin  Panel  Concluding Panel Michel Dagenais, Robert Charpentier and Dominique Toupin will animate a discussion to identify the most important current and upcoming challenges for system analysis shared by the participants.