add 3rd gen viewer architecture document

[lttv.git] / trunk / masters-pmf / 3rd_gen_viewer / 3rd_gen_viewer_arch.lyx

#LyX 1.5.5 created this file. For more info see http://www.lyx.org/
\lyxformat 276
\begin_document
\begin_header
\textclass article
\language english
\inputencoding auto
\font_roman default
\font_sans default
\font_typewriter default
\font_default_family default
\font_sc false
\font_osf false
\font_sf_scale 100
\font_tt_scale 100
\graphics default
\paperfontsize default
\spacing single
\papersize default
\use_geometry false
\use_amsmath 1
\use_esint 1
\cite_engine basic
\use_bibtopic false
\paperorientation portrait
\secnumdepth 3
\tocdepth 3
\paragraph_separation skip
\defskip medskip
\quotes_language english
\papercolumns 1
\papersides 1
\paperpagestyle default
\tracking_changes false
\output_changes false
\author "" 
\author "" 
\end_header

\begin_body

\begin_layout Title
A Set of APIs for a Third-Generation Trace Viewer
\end_layout

\begin_layout Author
\noindent
Pierre-Marc Fournier, Michel Dagenais, Mathieu Desnoyers
\newline

\family typewriter
\size footnotesize
<pierre-marc.fournier _at_ polymtl.ca>
\family default
\size default

\newline

\newline
CASI Lab
\newline
École Polytechnique de Montréal
\newline

\newline
WORK IN PROGRESS
\end_layout

\begin_layout Section
Introduction
\end_layout

\begin_layout Standard
This document proposes some APIs for a trace viewing and analysis infrastructure.
\end_layout

\begin_layout Standard
Design choices were made based on the experience acquired by the design,
 development and use of two previous generations of trace viewers, respectively
 the Trace Visualizer from the Linux Trace Toolkit and its successor, the
 LTTV from the LTTng tracing toolkit.
\end_layout

\begin_layout Section
Goals
\end_layout

\begin_layout Itemize
Provide an infrastructure for fast, interactive visualization and analysis
 of huge traces (>10 GB) on standard workstations
\begin_inset Foot
status open

\begin_layout Standard
Efficiency is of great importance.
 It is tempting to think that as computers get faster, it will be possible
 to run more complex analyses on traces in a shorter time.
 In fact, the size of the traces grows as computers get more powerful, because
 they do more operations than before in the same amount of time and therefore
 generate more events.
 The current tendency to parallelize systems is another factor that results
 in bigger traces.
\end_layout

\end_inset


\end_layout

\begin_layout Itemize
Support live trace processing
\end_layout

\begin_layout Itemize
Allow merging of traces of different formats
\end_layout

\begin_layout Standard
\begin_inset Note Note
status open

\begin_layout Itemize
Design the viewer in such a way that it can run in parallel on multi-core
 systems 
\begin_inset Note Greyedout
status open

\begin_layout Standard
(Is it worth it? Beware of complex locking issues)
\end_layout

\end_inset

; No, we don't.
\end_layout

\end_inset


\end_layout

\begin_layout Itemize
Design with both command line utilities (C language) and graphical interfaces
 (Java and others) in mind, sharing code and standardizing interfaces as
 much as possible
\end_layout

\begin_deeper
\begin_layout Standard
Both command-line interfaces and graphical interfaces have an important
 role to play in trace analysis.
 While graphical interfaces allow for complex views, command-line utilities
 can be used quickly in more environments; they can also be more convenient
 in some circumstances.
 A good design should be such that they can share code and APIs.
\end_layout

\end_deeper
\begin_layout Section
General aspects
\end_layout

\begin_layout Itemize
All data structures must be accessed through API methods, in order to allow
 for internal change without breaking compatibility.
\end_layout

\begin_layout Section
Controlling the tracing of a system
\end_layout

\begin_layout Subsection
Overview
\end_layout

\begin_layout Standard
The goal of this API is to provide a portable way of controlling tracing
 of heterogenous systems.
\end_layout

\begin_layout Standard
It should enable to do the following actions.
\end_layout

\begin_layout Itemize
Set the parameters of a trace (channels, buffering, destination of data
 (file, network, process)...) 
\end_layout

\begin_layout Itemize
Control the recording of events (start, stop, pause the trace)
\end_layout

\begin_layout Itemize
Create tracepoints (on dynamic tracers) and control which tracepoints are
 activated (some may be at trace level, others might only permit system
 level)
\end_layout

\begin_layout Subsection
Trace control API
\end_layout

\begin_layout Itemize
List the static tracepoints available on a system
\end_layout

\begin_deeper
\begin_layout Standard
These may be static tracepoints (active or inactive) or dynamic tracepoints
 (active or proposed).
\end_layout

\end_deeper
\begin_layout Itemize
Add a new dynamic tracepoint
\end_layout

\begin_layout Itemize
Activate a tracepoint
\end_layout

\begin_layout Itemize
Deactivate a tracepoint
\end_layout

\begin_layout Itemize
List available probes
\end_layout

\begin_layout Itemize
Connect a probe to a tracepoint
\end_layout

\begin_layout Itemize
Start tracing
\end_layout

\begin_layout Itemize
Stop tracing
\end_layout

\begin_layout Itemize
\begin_inset Note Greyedout
status open

\begin_layout Standard
others missing...
\end_layout

\end_inset


\end_layout

\begin_layout Section
Reading traces
\end_layout

\begin_layout Subsection
Overview
\end_layout

\begin_layout Subsubsection
Timestamps
\end_layout

\begin_layout Standard
Timestamps, at the framework level, should be in an architecture-independent
 format.
 One possibility would be to have them stored in seconds (with decimals).
 Offsetting, if necessary, could be handled by the traceset (trace merging)
 level.
 An uncertainty value should accompany the timestamp.
 Timestamps in cycles should be available if they are provided by the tracer.
\end_layout

\begin_layout Subsection
Low-level trace reading API
\end_layout

\begin_layout Standard
The goal of this API is to provide a uniform way for the framework to access
 the events recorded in a trace, in an encoding-independant way.
\end_layout

\begin_layout Subsubsection
Factory methods
\end_layout

\begin_layout Itemize
get_supported_formats()
\end_layout

\begin_deeper
\begin_layout Standard
Arguments: none
\end_layout

\begin_layout Standard
Return:
\end_layout

\begin_layout Itemize
a list of supported formats
\end_layout

\end_deeper
\begin_layout Itemize
open_trace()
\end_layout

\begin_deeper
\begin_layout Standard
Arguments:
\end_layout

\begin_layout Itemize
trace URL
\end_layout

\begin_deeper
\begin_layout Itemize
local file name
\end_layout

\begin_layout Itemize
other special url to establish a network connection
\end_layout

\begin_layout Itemize
etc...
\end_layout

\end_deeper
\begin_layout Itemize
trace format (optional, for cases where auto detection would not work or
 be ambiguous)
\end_layout

\begin_layout Standard
Return:
\end_layout

\begin_layout Itemize
a trace handle
\end_layout

\end_deeper
\begin_layout Subsubsection
Trace methods
\end_layout

\begin_layout Itemize
close()
\end_layout

\begin_layout Itemize
current_event()
\end_layout

\begin_deeper
\begin_layout Standard
Arguments: none
\end_layout

\begin_layout Standard
Return:
\end_layout

\begin_layout Itemize
the event currently pointed by the position pointer
\end_layout

\end_deeper
\begin_layout Itemize
advance()
\end_layout

\begin_deeper
\begin_layout Standard
Advance the position pointer to the next event in the trace.
\end_layout

\begin_layout Standard
Arguments: none
\end_layout

\begin_layout Standard
Return values:
\end_layout

\begin_layout Itemize
The result:
\end_layout

\begin_deeper
\begin_layout Itemize
Success
\end_layout

\begin_layout Itemize
Final end of trace (end of a non-live trace or of a completed live trace)
\end_layout

\begin_layout Itemize
No more events for now (end of a still running live trace)
\end_layout

\end_deeper
\end_deeper
\begin_layout Itemize
seek_time()
\end_layout

\begin_deeper
\begin_layout Standard
Arguments:
\end_layout

\begin_layout Itemize
The time to seek to
\end_layout

\begin_layout Standard
Return: none
\end_layout

\end_deeper
\begin_layout Itemize
get_position_handle()
\end_layout

\begin_layout Itemize
seek_to_position_handle()
\end_layout

\begin_layout Itemize
destroy_position_handle()
\end_layout

\begin_layout Itemize
get_time_start()
\end_layout

\begin_deeper
\begin_layout Standard
Arguments: none
\end_layout

\begin_layout Standard
Return:
\end_layout

\begin_layout Itemize
the timestamp of the first event in the trace
\end_layout

\end_deeper
\begin_layout Itemize
get_time_end()
\end_layout

\begin_deeper
\begin_layout Standard
Arguments: none
\end_layout

\begin_layout Standard
Return:
\end_layout

\begin_layout Itemize
the timestamp of the last event in the trace
\end_layout

\end_deeper
\begin_layout Itemize
register_callback_new_event()
\end_layout

\begin_deeper
\begin_layout Standard
Register a callback that is called when a new event becomes available in
 a live trace.
 It is also called when the live trace ends.
\end_layout

\begin_layout Standard
Arguments:
\end_layout

\begin_layout Itemize
after (timestamp) : call only if the event occurent later than 
\emph on
after
\end_layout

\begin_layout Itemize
the callback function
\begin_inset Note Note
status open

\begin_layout Standard
specify its args and return val
\end_layout

\end_inset


\end_layout

\begin_layout Standard
Return:
\end_layout

\begin_layout Itemize
Result (success or failure)
\end_layout

\end_deeper
\begin_layout Subsection
High-level trace reading API
\end_layout

\begin_layout Standard
The goal of this API is to provide a uniform way for analyses and views
 to obtain trace events from a traceset (merge of many traces or a single
 trace), regardless of the system they were recorded on and of the tracer
 used.
\end_layout

\begin_layout Itemize
Request a range of events
\end_layout

\begin_deeper
\begin_layout Standard
Arguments:
\end_layout

\begin_layout Itemize
(implicitly) the traceset
\end_layout

\begin_layout Itemize
start timestamp
\end_layout

\begin_layout Itemize
stop timestamp (special value for infinity, for live traces)
\end_layout

\begin_layout Itemize
list of event types and callbacks
\end_layout

\begin_layout Itemize
filter with complex expressions
\end_layout

\begin_layout Itemize
private pointer to be passed to the callbacks
\end_layout

\begin_layout Standard
Return:
\end_layout

\begin_layout Itemize
handle to the request for cancelling it
\end_layout

\end_deeper
\begin_layout Itemize
Cancel a request
\end_layout

\begin_deeper
\begin_layout Standard
Arguments:
\end_layout

\begin_layout Itemize
the handle to the request
\end_layout

\end_deeper
\begin_layout Subsection
State management
\end_layout

\begin_layout Subsubsection
State accessing API
\end_layout

\begin_layout Itemize
Request the values of a given set of state variables at a point in time
\end_layout

\begin_deeper
\begin_layout Standard
Arguments:
\end_layout

\begin_layout Itemize
the list of state variables
\end_layout

\begin_layout Itemize
the timestamp
\end_layout

\end_deeper
\begin_layout Itemize
Request all the states changes of a given set of state variables between
 two time indexes
\end_layout

\begin_deeper
\begin_layout Standard
Arguments:
\end_layout

\begin_layout Itemize
the list of state variables
\end_layout

\begin_layout Itemize
timestamp 1
\end_layout

\begin_layout Itemize
timestamp 2
\end_layout

\end_deeper
\begin_layout Subsubsection
State setting API
\end_layout

\begin_layout Section
Describing event types
\end_layout

\begin_layout Subsection
Overview
\end_layout

\begin_layout Standard
Because tracepoints may be created dynamically, information about the descriptio
n of events is just as dynamic.
 In this context, one simple way to communicate the event description informatio
n to upper layers would be to send them as events, as it is done in recent
 versions of LTTV.
 The core events used to describe other events are the only ones whose descripti
on is hardcoded in the framework.
\end_layout

\begin_layout Standard
These event-type-describing events could then be received and interpreted
 by a module, the Event Description Service, which would be a client to
 the high-level tracing API at the same level as normal views and analyses.
 It would store the information and allow the other views and analyses to
 access it via this API.
\end_layout

\begin_layout Subsection
Events-describing events
\end_layout

\begin_layout Itemize
Event type declaration event
\end_layout

\begin_deeper
\begin_layout Standard
Announce the existence of an event type
\end_layout

\begin_layout Itemize
event type name
\end_layout

\end_deeper
\begin_layout Itemize
Argument declaration event
\end_layout

\begin_deeper
\begin_layout Standard
Announce the existence of an event argument
\end_layout

\begin_layout Itemize
event type name
\end_layout

\begin_layout Itemize
name
\end_layout

\begin_layout Itemize
type
\end_layout

\end_deeper
\begin_layout Itemize
Event destruction
\end_layout

\begin_deeper
\begin_layout Standard
Announce that an event type ceased to exist
\end_layout

\begin_layout Itemize
event type name
\end_layout

\end_deeper
\begin_layout Subsection
Event type description API
\end_layout

\begin_layout Itemize
Get the list of all the event types
\end_layout

\begin_layout Itemize
Find an event type by name
\end_layout

\begin_layout Itemize
Get the number of arguments of an event
\end_layout

\begin_layout Itemize
Get the list of arguments of an event
\end_layout

\begin_layout Itemize
Find an argument by name
\end_layout

\begin_layout Itemize
Get the type of an argument of an event type
\end_layout

\begin_layout Itemize
Get a string representation of an argument of any type, given an event that
 contains it an instance of it
\end_layout

\begin_layout Itemize
Get an integer representation of an integer argument, a floating-point represent
ation of a floating-point argument
\end_layout

\begin_layout Itemize
Functions for accessing other types
\end_layout

\begin_layout Section
Inspecting events
\end_layout

\begin_layout Subsection
Overview
\end_layout

\begin_layout Standard
Events contain the following information.
\end_layout

\begin_layout Itemize
Timestamp
\end_layout

\begin_layout Itemize
Event type identifier
\end_layout

\begin_deeper
\begin_layout Standard
Two possibilities:
\end_layout

\begin_layout Itemize
an event id (integer)
\end_layout

\begin_layout Itemize
an event name (string)
\end_layout

\end_deeper
\begin_layout Itemize
The name of the machine it occured on
\end_layout

\begin_layout Subsection
Event inspecting API
\end_layout

\begin_layout Itemize
get the event type identifier
\end_layout

\begin_layout Itemize
get timestamp
\end_layout

\begin_layout Itemize
get the name of the machine on which the event occured
\end_layout

\begin_layout Itemize
get information on the type of tracing technology that was used
\end_layout

\begin_layout Itemize
get the corresponding tracepoint (machine/tracing technology/name/location
 in code(if available))
\end_layout

\end_body
\end_document