TotalView

1
TotalView
2
Introduction
TotalView Technologies products set the standard
for the interactive analysis and debugging of
serial and parallel codes.

• TotalView Debugger
• Comprehensive debugging solution that
  dramatically enhances and simplifies the process
  of debugging parallel, data-intensive,
  multi-process, multi-threaded or
  network-distributed applications
• MemoryScape Memory Debugger
• Easy-to-use, graphical, interactive memory
  debugger to help inspect and resolve difficult
  memory problems including complex multi-process
  and multi-threaded programs
• ReplayEngine
• Freedom to move forward or backwards through
  execution history, and work in reverse from from
  program crashes

3
TotalView Debugger
Value Proposition

• TotalView enhances platforms to be more
  productive development environments
• Accelerates the pace of research and the time to
  delivery for products and solutions
• Efficiently locates bugs in data-intensive,
  multi-process, multi-threaded, large clusters,
  parallel processing environments
• Reduces development time to enable quicker
  implementation for parallel and distributed
  applications.
• Allows for more cost effective and efficient
  development and provides a rapid return on
  investment (ROI)

4
TotalView Debugger Highlights

• Great for Threads
• Race conditions can take large engineering effort
  to pin down and resolve
• Control over threads allows recreation of race
  conditions
• TVD Debugs OpenMP and Hybrid OpenMP/MPI
• TVD helps development teams take the risk out of
  multithreaded development
• Essential for Clusters
• Work with many processes with as little as two
  windows
• Operations can apply narrowly or broadly
• Scalable up to thousands of processes

5
TotalView Debugger Highlights

• Ease of Use for Beginners
• Common operations on the GUI
• just click to get more information about
• Variables
• Functions
• Breakpoints
• Processes
• Highly Graphical
• Visualize array data
• A picture is worth 1000 words
• Memory data displayed graphically (I.e, red
  blocks are leaks)
• Call graph of program structure
• Message queue graph

6
TotalView Debugger Highlights

• Memory Debugging
• Easy to apply -- no recompilation or
  instrumentation
• Lightweight -- minimal impact on application
  runtime
• Powerful -- points to leaks before they become
  floods!
• Interactive graphical display
• Makes the connection for the developer between
  memory and code
• Cross Platform
• Project and code lifespan is often longer than
  the lifespan of a hardware platform
• TVD has very very similar functionality on a wide
  range of systems
• Developers and scientists can leverage investment
  in learning TotalView across their careers
• Support for all kinds of jobs workstations,
  servers, clusters and supercomputers

7
What Is TotalView?
8
Supported Compilers, Distros and Architectures
Product Overview TotalView Debugger

• Platform Support
• SGI IRIX
• Linux x86, x86-64, ia64, Power
• Mac Power and Intel
• Solaris Sparc and AMD64
• HPUX, IA64, AIX, Tru64, IRIX
• Cray X1, XT3, IBM BGL, SiCortex
• Languages / Compilers
• C/C, Fortran, UPC, Assembly
• Many Commercial Open Source Compilers
• Parallel Environments
• MPI (MPICH1 2, LAM, Open MPI, poe, SGI MPT,
  Quadrics, MVAPICH, HP MPI many others )
• UPC (HP INTREPID)

9
Multi-Thread Debugger
Product Overview TotalView Debugger

• Threads
• Easily understand which thread you are looking at
• Quickly switch thread context
• Monitor the state of all threads
• View data
• Shared
• Stack
• Thread Private
• Control Individual Threads
• Synchronize
• Run/Step
• Hold

10
Asynchronous Process Thread Control
Product Overview TotalView Debugger

• Focus on just one or a few processes or threads
• Visibility of what is happening in whole job
• Switch focus easily
• Control at process or thread level
• Asynchronous Control
• Step, Go/Halt
• Hold
• Control with groups of threads or processes
• Go/Halt
• Step
• Synchronize processes or threads
• Breakpoints
• Run-To

11
MPI Debugging
Product Overview TotalView Debugger

• Automatic Process Acquisition
• Efficient Cluster Architecture
• Advanced Process Control
• View Parallel Data
• Parallel Expression Evaluation
• Message Queue Display
• Highly Scalable

12
Memory Debugger Features

• Highly Graphical Interface
• Heap Memory Visualization
• Automatic Leak Detection
• Dangling pointer Indication
• Memory Corruption Tracking-Configurable Guard
  Blocks
• Memory Comparisons between Processes
• Also available standalone, as MemoryScape

13

• New Product Introduction
• ReplayEngine

14
TotalViewNew Product Introduction

• ReplayEngine
• Until now, debugging always meant running only
  forward to (hopefully) reach your problem
• But unfortunately, just one step too far often
  means having to start over again from the
  beginning, and hoping the program behavior will
  be the same!
• With ReplayEngine, you can simply unstep back
  through the program and seek out its cause -
  deterministically!
• Replay gives you a second chance to see what
  happened earlier in your program - and without
  starting over.

15
TotalViewNew Product Introduction

• ReplayEngine
• Captures full execution history
• Record all external input
• Record other sources of non-determinism
• Replay execution
• Replay the execution along exactly the same
  trajectory
• Can reset to any point along the path already
  taken
• Everything is managed by the tool
• The user just chooses where to go
• Back into or over functions, one line, or to the
  beginning
• There is some overhead
• Lots of clever tricks are used to reduce overhead
• Provides Determinism within a debugging session

16
TotalViewNew Product Introduction

• Navigation with ReplayEngine
• - an Intuitive User Interface -

Step forward over functions
Step backward over functions
Step forward into functions
Step backward into functions
Advance forward out of current Function, after
the call
Advance backward out of current Function, to
before the call
Advance forward to selected line
Advance backward to selected line
Advance forward to live session
17
TotalViewNew Product Introduction

• ReplayEngine

New Toolbar
No Replay? (or not running)
No Active Buttons!
18
TotalViewNew Product Introduction

• ReplayEngine
• Consider the following very difficult program
  scenario
• A crash occurs that destroys the stack backtrace,
  giving no information leading up to the problem
• ReplayEngine can be used to work backwards from
  the crash, and even to observe the stack recreate
  itself, providing the critical information on
  where and how the problem began.
• The ReplayEngine provides the ability to review
  any part of the program execution to see all
  variables and function calls, from the beginning
  of the run to the current time

19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22
(No Transcript)
23
(No Transcript)
24
ReplayEngine Scenarios

• Especially useful in scenarios that include
• Any crash that destroys the stack backtrace
• Race conditions with multi-threaded programs
• Server applications responding to network input
• Hard-to-reproduce GUI errors
• Crashes with invalid pointers
• Crashes involving write-after-free errors
• Examining complex functions and subroutines
• Writing past the end of an array
• Errors that depend sensitively on input
• We see these kinds of scenarios in
• Commercial applications
• Technical computing
• Education
• Scientific programming
• Network programming

25
Why Reverse Debugging?

• Thats the direction you need to go
• Backwards, from the effect to the cause
• Drastically reduces the effort
• No need to restart and run over again from the
  beginning
• Lets you focus
• Ensures reproducibility
• No chance that youll miss something and not be
  able to get it back
• Race conditions
• Crashes
• Input-dependent behavior

26
ReplayEngine Customer Feedback

• The best thing about ReplayEngine is being able
  to step backward. The implementation is done in a
  such a manner that it's intuitive to use.
• Recently we used the ReplayEngine to solve 2
  bugs to identify a wrong input and to find an
  error into an algorithm. We had a good feeling
  about the experience because the engine is
  simple to use and although it's a bit slow it
  allows us to go back without trying to guess
  the right breakpoint and restart the
  application.
• ReplayEngine is quite seamlessly integrated
  with the TotalView Debugger.
• Being able to change breakpoints during replay
  is a great feature.
• This will improve my productivity because I
  don't have to repeatedly restart the program to
  get to the execution point to examine the
  execution state.
• I would tell my colleagues that this is a great
  tool
• ReplayEngine...allows me to work backwards from
  the program crash toward the defect.  This is
  much nicer than having to repeatedly set
  breakpoints and restart my program from the
  beginning. - Dan Hiepler, Senior Developer,
  Intelligent Light, Inc.

27
TotalViewNew Product Introduction
ReplayEngine

• Currently available on Linux-86 and Linux-86-64
  machines.
• Current MPI Support
• HP MPI
• OpenMPI
• MVAPICH2
• Intel MPI
• MPICH 2
• LAM

28
Whats Newin TotalView 8.6
29
TotalView 8.6

• Non-Interactive Debugging
• tvscript provides non-interactive batch-mode
  debugging with TotalView
• Useful for long jobs or jobs larger than
  interactive queue
• More powerful and flexible than Print-style
  debugging
• Can be used to automate test/verify environments
• Leverages full power of TotalView in a batch mode
• Utilizes a simple Event/Action notation
• Provides a single file with clean, formatted
  output

Think of tvscript as Printf on steroids!
29
30
TotalView 8.6

• Remote Display Client
• The Remote Display Client offers users the
  ability to easily set up and operate a TotalView
  debug session that is running on another system.
• TotalView Remote Display has two components
• a Client, which runs on the remote system
• a Server, which runs on any system supported by
  TotalView, invisibly manages the connections
  between the host and client
• The Remote Display Client is available for
• Linux x86
• Linux x86-64
• Windows XP
• Windows Vista

Remote TotalView Debugging thats Easy, Secure
and Fast
31
TotalView 8.6
32
TotalView 8.6

• Remote Display Security
• Remote Display uses SSH
• Remote Display Server allows only RFB (Remote
  Frame Buffer) connections to and from the host
• No incoming access to the Server is allowed
• The Server can only connect back to the Viewer
  via SSH
• Only one Viewer connection is allowed to the
  Server
• As Remote Display connects to systems, the user
  is prompted for password information as required
• Also required
• ssh must be available on all nonWindows systems
  being accessed.
• X Windows must be available on the remote system.
  

33
TotalView Technologies

• Contato
• Alfredo Leon
• alfredo.leon_at_tecpartner.com.br
• Ariani Félix
• ariani.felix_at_tecpartner.com.br
• Telefones (11)5183-6579
• (11)5183-4066