Menu
▾
▴
[flexotask-commits] SF.net SVN: flexotask:[107] trunk/flexotask/doc/flexotaskProgramming.html
|
From: <jsa...@us...> - 2008-11-17 17:05:52
|
Revision: 107
http://flexotask.svn.sourceforge.net/flexotask/?rev=107&view=rev
Author: jsauerbach
Date: 2008-11-17 17:05:46 +0000 (Mon, 17 Nov 2008)
Log Message:
-----------
Fill in some sections that are just fast-edits of stuff in the old Exotask document
Modified Paths:
--------------
trunk/flexotask/doc/flexotaskProgramming.html
Modified: trunk/flexotask/doc/flexotaskProgramming.html
===================================================================
--- trunk/flexotask/doc/flexotaskProgramming.html 2008-11-17 17:05:08 UTC (rev 106)
+++ trunk/flexotask/doc/flexotaskProgramming.html 2008-11-17 17:05:46 UTC (rev 107)
@@ -1343,19 +1343,110 @@
<br><hr><h3><a name="tasks"></a>Writing Flexotasks</h3>
-<p>Under construction.
+<p>This section is still being written. For the moment you are
+referred to the <a href="#samples">sample programs</a> for ideas about what
+the actual code of Flexotasks should look like.
<br><hr><h3><a name="validation"/>Validating and Instantiating Flexible Task Graphs</h3>
-<p>Under construction.
+The <a
+href="com/ibm/realtime/flexotask/template/FlexotaskTemplate.html">FlexotaskTemplate</a>
+class provides five different validate methods providing convenient
+combinations of four arguments, of which only one argument (the
+<b>schedulerKind</b>) is always required. Here is a description of
+what the arguments do.
+<ul>
+<li><b>schedulerKind--</b> String name of the kind of scheduler to use
+in running the graph. The <b>TTScheduler</b> (time triggered
+scheduler) is always available. Other schedulers are provided by
+plugins (which then also contribute to the generated runtime). This
+is the only "required" argument and is present in all five
+<b>validate</b> methods.
+
+<li><b>schedulingData--</b> conveys platform-specific worst case
+execution time (WCET) and worst case allocation (WCA) information for
+use by the scheduler, as discussed in the next section. If this
+argument is omitted, the scheduler assumes default heap sizes for
+Flexotask private heaps and assumes that the worst case execution
+times of tasks are short enough to make the schedule feasible without
+further checking. This argument is optional. The <b>TTScheduler</b>
+only looks at WCA information and ignores WCET information. The WCA
+information is used to estimate heap sizes and garbage collection
+strategies for the private memory areas.
+
+<li><b>initializationMap--</b> provides values to be passed as the
+parameter arguments to one or more Flexotasks' <b>initialize</b>
+methods. If those Flexotasks already have default parameter values
+specified as part of the template, the defaults are overridden by the
+contents of <b>initializationMap</b>. Weakly isolated Flexotasks
+typically use dynamically specified parameters to communicate with the
+rest of the Java application, as in the <b>communicating</b> sample
+program. Strongly isolated Flexotasks receive deep copies of their
+parameters. If this argument is missing it is assumed to be the empty
+map.
+
+<li><b>distributionContext--</b> is present only in the most general
+of the five <b>validate</b> methods. It provides information that can
+be used to distribute the graph across multiple machines. If missing,
+the graph is assumed not to be distributed. Distributed graphs
+require the presence of a system extension called a
+<em>distributer</em>. The present system does not contain any
+distributers.
+</ul>
+
<br><hr><h3><a name="wcet"/>Specifying WCET and WCA information</h3>
-<p>Under construction.
+<p>The <a
+href="com/ibm/realtime/flexotask/scheduling/FlexotaskSchedulingData.html">FlexotaskSchedulingData</a>
+data structure is used to pass platform-specific WCET and WCA
+information to the scheduler. It provides a scheduler-independent
+framework for specifying this information, but individual schedulers
+may extend the class to provide scheduler-specific addenda or a more
+convenient interface given the properties of the scheduler. The
+scheduling data is expressed as a map from FlexotaskTemplate elements
+(task or connection) to <a
+href="com/ibm/realtime/flexotask/scheduling/FlexotaskTaskSchedulingElement.html">FlexotaskTaskSchedulingElement</a>
+and <a
+href="com/ibm/realtime/flexotask/scheduling/FlexotaskConnectionSchedulingElement.html">FlexotaskConnectionSchedulingElement</a>
+structures, with the addition of an <a
+href="com/ibm/realtime/flexotask/scheduling/FlexotaskGlobalSchedulingElement.html">FlexotaskGlobalSchedulingElement</a>
+structure. Tasks have WCETs both for execution and for garbage
+collection, and have WCAs, while connections have only WCETs for their
+deep cloning operations. The global scheduling element provides an
+initial and maximum total aggregate size for Flexotask private heaps
+and also specifies the number of cores available on the target
+platform.
+<p>The TTScheduler only considers WCA and heap size information. It
+makes no use of WCETs or information about the number of cores. Other
+schedulers are being developed that will use some more of this
+information.
+
<br><hr><h3><a name="cloning"/>Deep cloning at test time</h3>
-<p>Under construction.
+<p>When testing a Flexotask graph (not running on a real-time VM), the
+Flexotask runtime must simulate the deep cloning operation that is
+done specially in the real-time VM at the system level. It uses Java
+level (including reflective) operations to do this, which is not
+ideal, since Java reflection does not provide a consistent
+side-effect-free way of constructing objects. The solution is to use
+the various implementations of the <a
+href="com/ibm/realtime/flexotask/cloning/DeepCloner.html">DeepCloner</a>
+interface provided in the <a
+href="com/ibm/realtime/flexotask/cloning/package-summary.html">com.ibm.realtime.flexotasks.cloning</a>
+package, or create one of your own. By default, the system uses
+either the <b>SerializableCloner</b> (for objects that implement
+Serializable) or the <b>DefaultCloner</b> (otherwise). The latter
+assumes that the object implements a <b>deepClone()</b> method, which
+is fine when you are creating the data types anew for the application,
+but not so good when you are using existing types. To create an
+association between a <b>Class</b> and its <b>DeepCloner</b> use the
+<a
+href="com/ibm/realtime/flexotask/cloning/CloningSupport.html#setCloner(java.lang.Class,%20com.ibm.realtime.flexotask.cloning.DeepCloner)">CloningSupport.setCloner()</a>
+method. Note that reflective cloning is not used when executing in
+real-time mode, so the performance of your cloning solution is not an
+issue in determining WCETs for connections.
<br><hr><h3><a name="tracing">Using the Tracing Support</h3>
This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.
|
